Bob Cordell Interview: BJT vs. MOSFET

darkfenriz said:
Bob
What is your opinion of an output stage with bjt driving hexfet driving bjt?
All complementary followers.
Like:
2n5401/2n5551 for pre-drivers
IRF610/9610 for drivers (with ~80mA idle current)
and 2sa1943/2sc5200 for power devices?

regards
Adam


Hi Adam,

I'm sure it could be made to work well if properly employed, but what did you have in mind?

If I was going to do a bipolar output, I could be tempted to use a MOSFET to drive the output transistor, but there might be some concern about the bias stability of the overall architecture if one was driving an emitter follower bipolar output with a source follower MOSFET.

In my amps, I generally like a N-channel JFET input differential pair, a bipolar signal path in the middle, and a HEXFET output stage. I do use MOSFET current sources to drive my folded emitter follower bipolar drivers.

Cheers,
Bob
 

GK

Disabled Account
Joined 2006
Bob Cordell said:
Glenn,

Please tell me what it is about Figure 4 in my MOSFET amplifier paper that you do not understand.

Bob



I don't think there is anything about Figure 4 in your MOSFET amplifier paper I do not understand. You seem to think it shows something that conflicts with something specific I said. Care to elaborate?
 
john curl said:
Personally, I would tend to choose the 2sk1530 over the IRF240. What would be wrong with that?


Hi John,

This looks like a pretty good part. It is a vertical MOSFET, and appears to have a somwhat lower Vgs for a given current than the IRFP240. Capacitances and gm vs Id are similar. Note the significant voltage dependence of the Co and Cr. The TCvgs crossover current is about 8 amps, a bit lower than that for the 240, possibly suggesting slighly better thermal stability. The 10 ms SOA at 100V appears to be 10 amps, unless I've read it wrong. This is VERY good.

At a price of $8 @ 100, (as opposed to $1.80 for IRFP240) one does need to consider carefully whether it brings that much more to the table than the IRFP240 in a typical application. It would be interesting to see how much of the larger SOA is real as opposed to degree of conservatism in specifying it (but this doesn't change the fact that my IRFP240 devices did not make it to 10 amps at 100V).

Cheers,
Bob
 
Bob Cordell said:
The lower gm of the MOSFETs significantly mitigates their 6 mV/C TCvgs

Ditto the much higher treshold voltage - important when one takes into account that the usual bias arrangement is a 'voltage source'. If one uses a Vbe multiplier, one must multiply the Vbe by a factor of roughly 8 for two HEXFETs, which already gives 4x Vbe Tc per each HEXFET - aboout -8.8mV/C, too much. A straight Vbe multipler will overcompensate - more so if there are source resistors. On the other hand, using a Vgs multiplier with a HEXFET will give nearly perfect tracking, since it's the exact same MOSFET cells doing the thermal sensing. Also, modulo Tc becomes lower for higher drain currents. As it is more than likely the Vgs multiplier HEXFET will run at lower than output idle current, there is an extra 'edge' to compensate even MOSFETs without source resistors.


We also have at least two kinds of thermal bias stability. First, there is passive thermal bias stability...

IT should also be mentioned that with HEXFETs there is really no such thing as optimum bias, as gm is low and transfer characteristic follows a different rule than with BJTs. Although this can be viewed as a fault, at least in terms of thermal stability versus performance, it is not as important to get the bias current absolutely constant with temperature.
On the other hand, with modern array technology BJTs (LAPT, ring emitter, etc), there are internal emiter resistances, so figuring an 'ideal' idle current based on external Re voltage drop at idle is not a reliable method.

MOSFETs like to have a decent amount of drive current available for both turn-on and turn-off. In many cases this is less than what a bipolar needs, but it certainly does not obviate the need for adequate buffering between the VAS and the output transistors.

The delta(Ib) portion of a bipolar transistor Cin is typically smaller (easily by order of magnitude) compared to the required Ib. This means that with a buffer stage (which would be required for BJT outputs), the Vas sees a more constant impedance to drive - there is essentially the input capacitance in parallel with a scaled up load impedance.
With MOSFETs, delta(Ig) is actually all that is there, and in fact it is scaled up as load decreases (assuming a MOSFET follower, apparent Cin rises as load falls due to higher delta(Vgs) required for higher delta(Id)). The VAS sees only a capacitive load.
It may be argued that modern BJTs which are also constructed as arrays, have signifficant Cin, on the order of MOSFETs - the difference is, the requirement for a driver stage scales it down for the VAS. Not so if a 'VAS' is driving a MOSFET output directly (as is often the case). In order to make things equal, such a VAS would have to have an inordinately higher standing current - about (beta of BJT driver) higher than in the BJT output version! That is only the beginning of trouble - consider the phase shift and slew rate on the VAS output node, when it is so heavily loaded.
Things however turn favorable for a MOSFET as soon as there is clipping. Driving a MOSFET to saturation produces only as much charge storage as there is extra Vgs, and it is removed equally easily as it is put there. With BJTs, charge storage issues are signifficantly more difficult.
 
ilimzn said:


Ditto the much higher treshold voltage - important when one takes into account that the usual bias arrangement is a 'voltage source'. If one uses a Vbe multiplier, one must multiply the Vbe by a factor of roughly 8 for two HEXFETs, which already gives 4x Vbe Tc per each HEXFET - aboout -8.8mV/C, too much. A straight Vbe multipler will overcompensate - more so if there are source resistors. On the other hand, using a Vgs multiplier with a HEXFET will give nearly perfect tracking, since it's the exact same MOSFET cells doing the thermal sensing. Also, modulo Tc becomes lower for higher drain currents. As it is more than likely the Vgs multiplier HEXFET will run at lower than output idle current, there is an extra 'edge' to compensate even MOSFETs without source resistors.



IT should also be mentioned that with HEXFETs there is really no such thing as optimum bias, as gm is low and transfer characteristic follows a different rule than with BJTs. Although this can be viewed as a fault, at least in terms of thermal stability versus performance, it is not as important to get the bias current absolutely constant with temperature.
On the other hand, with modern array technology BJTs (LAPT, ring emitter, etc), there are internal emiter resistances, so figuring an 'ideal' idle current based on external Re voltage drop at idle is not a reliable method.



The delta(Ib) portion of a bipolar transistor Cin is typically smaller (easily by order of magnitude) compared to the required Ib. This means that with a buffer stage (which would be required for BJT outputs), the Vas sees a more constant impedance to drive - there is essentially the input capacitance in parallel with a scaled up load impedance.
With MOSFETs, delta(Ig) is actually all that is there, and in fact it is scaled up as load decreases (assuming a MOSFET follower, apparent Cin rises as load falls due to higher delta(Vgs) required for higher delta(Id)). The VAS sees only a capacitive load.
It may be argued that modern BJTs which are also constructed as arrays, have signifficant Cin, on the order of MOSFETs - the difference is, the requirement for a driver stage scales it down for the VAS. Not so if a 'VAS' is driving a MOSFET output directly (as is often the case). In order to make things equal, such a VAS would have to have an inordinately higher standing current - about (beta of BJT driver) higher than in the BJT output version! That is only the beginning of trouble - consider the phase shift and slew rate on the VAS output node, when it is so heavily loaded.
Things however turn favorable for a MOSFET as soon as there is clipping. Driving a MOSFET to saturation produces only as much charge storage as there is extra Vgs, and it is removed equally easily as it is put there. With BJTs, charge storage issues are signifficantly more difficult.


ilimzn,

These are very good points.

In my MOSFET power amplifier (posted on my site at www.cordellaudio.com under Published Papers), the EC complementary differential transistor pair forms the vbe multiplier. I placed one of the these two EC transistors on the heatsink, and one not on the heatsink, and this gave pretty good thermal bias compensation.

You are also right about optimum bias for the MOSFETs. They really don't get gm high enough, typically, near the crossover region to suffer from the so-called gm doubling effect.

The effective internal ballasting emitter resistance on the 2SC3264 ring emitter transistor is approximately 0.05 ohm, so, yes, this will add some uncertainty in biasing, but it should not be difficult to take into account.

Yes, the Ccb of these ring emitter transistors is every bit as much as the Cdg of a HEXFET, and just as voltage-dependent.

In my view, only a fool would drive a MOSFET output directly from an un-buffered VAS.

Thanks for your observations,
Bob
 
i suspect bob might have been generalizing a bit again. :)

we all know nelson biases his mosfets hotter than (most) conventional designs.

mlloyd1
PS - man o man, these are some GREAT discussions. reminds me of the old days of netnews before the web and AOL. i just wish charles hansen was up to joining in ...

Nelson Pass said:

Well Bob, that would be me.
:cool:
 
Nelson Pass said:


Well Bob, that would be me.

:cool:


Boy, did I goof!

Well, I know you are not a fool, that's for sure. So I guess I'm wrong, or at least I was way too quick to make a generalization.

My apologies.

OK, so what is your reasoning? Is your avoidance of the VAS buffer just confined to a particular topology? I'm trying to figure out how far apart we are on this, and under what conditions we agree/disagree.

Cheers,
Bob
 
john curl said:
Bob, why should we use mosfets, IF we have to drive them with buffers as well? It's not an improvement over bipolars.

I am not Bob, but when I add mosfet source follower in parallel to degraded emitter follower I can pump much more current per case without degradation of sonic qualities. I.e. I use 2 cases up to 30A peak total instead of 10 of them.
 
The one and only
Joined 2001
Paid Member
My apologies for embarrassing you, it's just the :devilr: in me.

Of course we agree that if we want a fast circuit whose
distortion remains low at high frequencies we need to toss
current into the Gates. I don't place as much emphasis on this
as you do, but I address it by running more current through
the Vas. In the X600.5 the symmetric voltage gain stage will
peak out at about 100 mA. We also take advantage of balanced
output stages (halving the slew requirement), and last but not
least, we bias the amplifiers high.

:cool:
 
john curl said:
Bob, why should we use mosfets, IF we have to drive them with buffers as well? It's not an improvement over bipolars.


John, that's a fair question. I've never considered that an advantage of using MOSFETs was any absence of a need to buffer the VAS for them.

So, clearly, whatever MOSFETs bring to the table as an improvement over bipolars, at least in my paradigm, must come even in the situation where there are predrivers and even drivers. Their benefit, if any, is not associated with a reduction in circuit complexity, at least in my designs.

One thing I like about them is their speed, both small-signal and the ability to turn them on and off.

Another thing I like about them is that the driver that they do need doesn't have to be as beefy, since there is no beta issue and no beta droop. For example, I don't think I could use folded emitter followers to drive bipolars; the standing current would have to be way too high in order to supply the worst-case base current.

I agree that in the absence of error correction, they are at a disadvantage in smaller amplifiers in that they need higher idle bias to achieve a given THD, at least in Class AB. In bigger amplifiers, where there tends to be more idle bias to go around anyway, I think this relative disadvantage pretty much goes away.

Although I have not had a chance to see real test-based SOA data on the ring emitter transistors, I think you have made a lot of progress in convincing me that the absence of second breakdown in the MOSFETs may not be as big an advantage over bipolars (at least the ring emitter devices) as I originally thought. At least on paper. I'm still very interested in seeing how well the ring emitter transistors hold up under a destructive test.

Bob
 
Nelson Pass said:
My apologies for embarrassing you, it's just the :devilr: in me.

Of course we agree that if we want a fast circuit whose
distortion remains low at high frequencies we need to toss
current into the Gates. I don't place as much emphasis on this
as you do, but I address it by running more current through
the Vas. In the X600.5 the symmetric voltage gain stage will
peak out at about 100 mA. We also take advantage of balanced
output stages (halving the slew requirement), and last but not
least, we bias the amplifiers high.

:cool:


Sounds good to me!

BTW, in your view is a balanced output stage different than a bridged output? A while back, bridging of amplifiers got a bad rap, it seems. Have we just changed the semantics to protect the innocent, or is there a substantive difference in architecture between an amplifier with a balanced output and one with bridged outputs? I'm assuming both have single-ended inputs, so that that part of the issue does not cloud the discussion.

Bob