Bob Cordell Interview: BJT vs. MOSFET

Hi Bob,

i know that J Curl and Leach favor bipolar output transistors as opposed to you favoring mosfets, care to elaborate on this?

again, your circuits seem to be single ended at input and VAS stages, am i correct or do you have any design that is fully complimetary from input to output?

thanks..
 
Why I prefer power MOSFETs

Tony said:
Hi Bob,

i know that J Curl and Leach favor bipolar output transistors as opposed to you favoring mosfets, care to elaborate on this?

again, your circuits seem to be single ended at input and VAS stages, am i correct or do you have any design that is fully complimetary from input to output?

thanks..


Hi Tony,

You have asked two very good questions. Both of these are likely to spark some discussion and maybe even some controversy, so I'll answer them in two separate posts. I'll begin here by dicussing my preference for power MOSFETs over bipolar output transistors.

I won't be able to go into all the details here, but there are two places on my web site at www.cordellaudio.com where most of the answers to this question lie. First, look under Power Amplifier Design/Why I Prefer Power MOSFETs. Secondly, look in Section 1 of my paper, "A MOSFET Power Amplifier with Error Correction". Although this paper is perhaps best known for its application of error correction to MOSFET output stages, a very big part of the paper is about the application and advantages of power MOSFETs to audio amplifier output stages.

As in most engineering decisions, there is a tradeoff between choosing MOSFETs versus bipolars for an output stage. Each technology has its advantages and disadvantages. You pick your poison. It is possible to make an outstanding amplifier with either technology, and it is also possible to make a terrible amplifier with either technology. The best amplifiers are made by those who know best how to deal with the limitations of the devices they choose.

Take a bipolar transistor and give it infinite beta at all currents and give it a Vbe of about 3.5V instead of the usual 0.7V. Give it about one-tenth the transconductance of a normal bipolar transistor at a given current. Make it 5 times as fast as a fast ring-emitter (sometimes called a perforated emitter) transistor. Make it have no secondary breakdown, with a failure mechanism that is purely thermal in nature. Give it a peak current capability in excess of 50 amps. Make its current as a function of bias voltage about ten times less sensitive to temperature than a normal bipolar transistor. You now have a power MOSFET. Admittedly, this is a crude approximation, but it helps draw the lines of distinction between the two devices.

I like MOSFETs because they are fast and immune to secondary breakdown. Their equivalent ft is on the order of 300 MHz. As a result, they produce very little dynamic crossover distortion. Protection circuits can be simple. They can source incredible current, and don't need a driver capable of sourcing high current. They have far superior thermal bias stability to bipolars, even though at normal bias current levels they also have a modest but positive temperature coefficient of current as a function of gate voltage.

They do have disadvantages. The most notable is their lower transconductance at a given current as compared to bipolars. This leads to higher output impedance in a source-follower arrangement, which can lead to higher static crossover distortion. This is one reason that they benefit so much from the application of error correction.

Power MOSFETs still make very good output stages even without error correction, but they need to be biased somewhat hotter than a bipolar stage to keep the static crossover distortion small. They work quite well with a standing bias current of, say, 200 mA, which will result in a source-follower output impedance on the order of one ohm per device, or 0.5 ohm for the complementary pair. Note that this results in an output stage idle power dissipation of about 24 watts with +/- 60V rails. Of course, if you're building a big honking power amplifier, there is nothing stopping you from running three or four pairs, each at 200 mA, to greatly reduce static crossover distortion while having a very nice Class AAB design.

People typically parallel numerous pairs of bipolar output transistors. This is for two reasons. First, it is to mitigate the beta droop problem at high currents. Secondly, it is to build up a larger total output stage Safe Operating Area so that less-intrusive protection circuits can be used. There is less need for this with power MOSFETs.

Keep in mind that you should match power MOSFETs if you are going to parallel them. This is pretty easy, however, since today's power MOSFETs from the same tube are usually like peas in a pod.

Cheers,
Bob
 
pavel:

i agree with you. however, isn't some of this capacitance effectively "mitigated" when the MOSFET is used in a source follower configuration?

mlloyd1

PMA said:
Bob,

are not you oversimplifying?

E.g., "do not need a driver capable of sourcing high current."

Really? In case input MOSFET capacitance is of some 1500pF, the driver has to deliver large

Ic = C*(dV/dt)

current, at least for fast signals.

Cheers,
Pavel
 
required current drive for MOSFETs

PMA said:
Bob,

are not you oversimplifying?

E.g., "do not need a driver capable of sourcing high current."

Really? In case input MOSFET capacitance is of some 1500pF, the driver has to deliver large

Ic = C*(dV/dt)

current, at least for fast signals.

Cheers,
Pavel


Pavel, this is a good question, but I was not really oversimplifying. This issue was covered in Section 1.5 of my MOSFET amplifier paper. The gate-source capacitance is bootstrapped by the output, so if the source follower gain is, say, 0.9, the 700 pF or so of Cgs is effectively reduced to about 70 pF. Then remains the gate-drain capacitance of about 100 pf. Together they amount to an effective capacitance of about 170 pF, thus requiring about 17 mA to support a voltage slew rate of 100 V/us.

Looking at the input capacitance of a device by itself can be very misleading. A lot of people express concern about the input capacitance of MOSFETs when they don't realize how big the effective input capacitance of a bipolar transistor is. Consider a ring emitter transistor with an ft of 30 MHz and operating at 1 amp. What do you think the effective input capacitance is? It is a whopping 0.2 uF! Just remember that the hybrid pi input capacitance of a bipolar transistor is transconductance in Siemens divided by 2*pi*ft. The transconductance at 1 amp is 40 S. Even though this may seem not to be a "physical" capacitance, it matters every bit as much. The thing that mitigates the effect of this capacitance is the same thing that mitigates it for a MOSFET: in an emitter follower configuration the hybrid pi capacitance is bootstrapped to a much smaller effective value.

What I really meant about high current drive required for bipolar output stages was that required under high output current conditions when there is beta droop in the output stage. In some cases, you can have an output stage delivering 10 amps to the load under conditions where the beta has drooped to 20 or less. In this case the driver needs to supply over 500 mA to the output devices. THIS is what I meant by high drive current. I've seen cases where the bipolar drivers go into secondary breakdown before the output transistors.

Cheers,
Bob
 
PMA said:
what do you think about 2SA1943/2SC5200 or their precedors 2SA1302/2SC3281, their beta is pretty linear up to some Ic = 8A.


What???

transistor_beta.gif
 
Lateral vs Vertical MOSFETs

Ultima Thule said:
Bob,

I guess your thoughts aplies mostly to vertical FET's, whats your take on lateral ones, not just paramters where the main differences like even lower transconductance, lower Cre and negative temp-co are the most noticeable but as well in practice?

Cheers Michael


Michael,

Much of what I said applies as well to the lateral power MOSFETs. There are many designs out there that use the laterals, and some people assert that they sound better. They are easier to use and are a more forgiving device than the vertical MOSFETs, but they don't perform as well as the verticals in my opinion.

The vertical MOSFETs are much faster, and some people have had bad experiences with the verticals when they did not exercise enough care in high frequency design and layout. Indeed, one of the few things that can cause a vertical MOSFET to destruct is a high-frequency parasitic oscillation, often above 100 MHz. The cause of the failure is the buildup of a.c. oscillation voltages internally on the gate in excess of 20 volts peak, which will punch through the oxide and destroy the device. This can happen even if the external gate-source voltage is clamped to be less than 20 volts as a result of the bondwire inductance working in resonance against the gate capacitance.

Although the vertical MOSFETs have significantly better temperature stability than bipolars, they are not as inherently temperature stable (bias-wise) as the laterals. The current at which the Id for fixed Vgs characteristic goes from a positive temperature coefficient (bad) to a negative coefficient (good) is around 100 mA for the laterals, while it is up around 1 amp for the verticals.

Cheers,
Bob
 
Wavebourn said:


Yes these transistors are quite good. Looks like you can get a worst-case beta of 50 at 7 amps per device (I use the typical low-temperature curve as worst-case room temperature). So if you parallel four devices, you can get 28 A into a pesky 2-ohm load, for a voltage of 56V peak. The total base current will then be about 560 mA, still not trivial, but certainly manageable. Still quite a bit more than would be required to drive a MOSFET at that level.

Under more forgiving conditions, you might put 60V peak into a 4-ohm resistive load, for 15A peak and an average power of 450 watts. Beta might be closer to 75. Peak base current would then be only 200 mA.

Bob
 
That is a good idea, Wavebourn. I was tempted to use something like that for a PA power amp. It has the problem of a mismatch in impedance, at both input and output, which could cause some real problems. Still, it can be an interesting approach.
Personally, I use FET's when they will allow the power output that I need to achieve. Most vertical FET's won't handle difficult loads or a short circuit, for more than an instant.
 
john curl said:
That is a good idea, Wavebourn. I was tempted to use something like that for a PA power amp. It has the problem of a mismatch in impedance, at both input and output, which could cause some real problems. Still, it can be an interesting approach.
Personally, I use FET's when they will allow the power output that I need to achieve. Most vertical FET's won't handle difficult loads or a short circuit, for more than an instant.

John, if you use real transistors you should not be afraid of impedance mismatches. Real problems happen when impedance mismatch happens when output current is crossing zero, and resulting voltage changes fine bias of output transistors. In my case it is nothing against 4V I have already, so guitar strings and reverberation fade out naturally, pleasing ears and soul. Yes, there are some crossover distortions, but they are not audible since power level when they happen is high enough. Also, input capacitances introduce lower impedance on higher frequencies, but suppose they are 200 pf effectively on background of 30 mA of base current... Nothing at all!
 
traderbam said:

One of the most unfortunate drawbacks for audio amps is the need for higher bias voltages and higher bias currents...combined these tend to make FET amps less power efficient.

...now let's compare it to filament and anode power losses of class A tube amps that sound better...

Russian joke: "What do you want really from taxi, painted chess-matt, or comfortable riding?"