Threshold voltage and transfer characteristics of lateral fet vs Hexfet

Hi all. I was just looking at a graph of the transfer characteristic of a 2SK134 mosfet and I noticed it was greatly different from a standard "switching" type Hexfet. Stick a drain voltage on the thing and as you start to wind up the gate voltage the drain current starts to climb almost from zero gate voltage. Contrast that with a Hexfet where nothing much happens until the gate voltage reaches 3-4 volts. Admittedly this is not an issue in their intended application, in fact it is sometimes a benefit because gate signals below a certain (noise) threshold are ignored, but for audio use this is a bit of a liability.

Can someone point me to somewhere that gives a comparison between these two types of fet, particularly for linear/audio usage?



2001-02-04 4:23 am
The K135/J50 pair of lateral MOSFET have about 1A per 1V gain, quite linear from about 1~7A. The Rds on is about 1R at room temperature rising to about 1R7 at high temperature. Your Circlotron would be much easier and stable to bias with lateral MOSFET. They do cost more though. The Semelab BUZ900D are the heaviest that I am aware of (while they list a Q version I have never seen one and I am sure they would want a ton of money for it based on the package it is in) at 16A/250W for about $14. The transition current from a bias standpoint is only 100mA on these. While the vertical IR type do show a similar transition around 3~5A, with any reasonable B+ they will be overloaded, and the gate tempco will negate the Rds tempco anyway. Nelson Pass is the only one I am aware of to successfully use the IR type in a linear amplifier. Constant temperature from massive heatsinks, class A bias (with a bias servo on some), and tight Vgs matching would seem to be his 'secret'. The failure of Counterpoint ('Time Bomb' and 'Grenade' come to mind when I hear the name Counterpoint) was in large part due to their inability to solve these problems.
I was only looking at those ones because I used them in a amp back in March '81. I wasn't actually considering using them. As far as what type of fets to use, I only want to use the Hex type because I have quite a few. And even then, I only want to make it with a circlotron topology. Remove those two factors and I dont even want an amp ;)

I get great satisfaction from putting smart feedback loops around pieces of dumb-iron, e.g. computerised engine management, and to make N-channel hexfets do what they are told is the whole purpose of the exercise. It's all part of the diy ethos. Reproducing music is perhaps secondary to the whole thing.

Nelson Pass is the only one I am aware of to successfully use the IR type in a linear

The IRF and other vertical mosfets have been used be several others including the the White Audio amps. Hundreds of DIY amplifiers have been built and are reliable. Bias stability is not hard to achieve using 1 ohm source resistors and or the usual bias compensation circuits. The reason the Counterpoint amps failed was that they used no source resisitors. Driving a short with a vertical mosfet without source resistors allows peak currents that are too high for the mosfet due to it's low on resistance. I believe this was the cause of the Counterpoint amp problems. Micheal Elliot of Counterpoint believed that the source resistors were audible and chose not to use them for that reason. A rather unfortunate choice I believe. I have played with this amp while replacing the filter caps and bias stability did not seem to be a problem. Lateral mosfets do not have high enough transconductance to make good output devices for power amps.

Maybe a little more research would be be in order before your next post on this subject.....

Lateral V. Hex mosfets


You raise interesting questions.

I have used the hexfets with great success in a Class A SE amplifier. They are extremely durable and thermally robust, and once their tendency to self-oscillation is manacled, they work very well.

You make the point that the high Vgs before they start to pass drain current is an issue for audio.

It is not really so much of a problem, however, since the bias at the complementary gates need only be adjusted to accommodate this extra voltage. What it does mean, however, is that quite a lot of voltage swing is lost from driver to output devices, which costs rail efficiency, and this is generally low even with lateral mosfets. The other difficulty is that the transconductance is not very high compared to bipolar devices, leading to a marked reduction in feedback factor, particularly at high output. Finally, there is that vexing issue of gate capacitance, the big one which is only solved with rigorous drive current settings.

There is a method of using the hexfet which largely overcomes these problems. We use an npn driver on the positive rail, and a pnp driver on the negative rail to drive the mosfets. These are respectively connected in complementary feedback configuration (cf. Sziklai pair). You require a p-type hexfet on the positive rail and an n-type hexfet on the negative rail. With six or seven mA through each driver, and the collector connected to the gate via a resistor from the rail, the emitters of each device is connected to the drain of its mosfet, just above the drain resistors which meet at the output. This gives strong gate drive with minimal swing at the gate - good for the gate capacitance issue.

Transconductance is greatly improved, giving essentially bipolar device feedback factors, and correspondingly, low distortion. The 100% local feedback effectively puts the 'mouse in charge of the elephant'. The topology also permits a conventional Vbe multiplier for bias control, and bias is now largely independent of the temperature of the output devices, which feature a positive tempco at high currents anyway. In effect, the transfer function of the stage is dominated by the almost linear characteristics of a small, fast bipolar device in almost constant current, and this is in sharp constrast to the conventional source follower beloved of audio designers around the world.

With any CFP there are issues of stability. A gate resistor of around 470R must be used to prevent charge shuttling back and forth to the gate and consequent oscillation. It is difficult to achieve unconditional stability in a Class AB design in my experience, but a Class A is usually not so difficult, and mosfets are ideally suited to Class A operation because of their exceptional thermal robustness.

FETs are like electrostatic speakers; they promise much and seem an elegant and purist solution to a range of problems, but in practice their tendency to self-oscillation makes them extremely difficult to use, particularly over they years as they age. In exasperation I gave up using them for audio almost three years ago!


Hugh R. Dean

I have tried the Szikali output stage in several amplifiers I built but usually had stability or thermal runaway problems and darlingtons seemed to work better for me.

I would like to know if you have tried putting the Szikali pair outside the feedback loop and if this would help the stability problem.

The Szikali pair always intrigued me but putting it into practice .......I found that the thermal problem was helped somewhat by makeing the bias circuit track the drivers and not the outputs.
Do you have any other thoughts on the matter?


Lateral V. Hex mosfets

"but in practice their tendency to self-oscillation makes them extremely difficult to use"

"A gate resistor of around 470R must be used"

Seems that you answered your own question. At the risk of starting another riot, repeat after... me phase margin, phase margin, phase margin...... Good phase margin makes for stable amplifiers.

If thermal stabilty of the bias current is a problem with some designs then would it not be possible to use DC-Servo to adjust the bias current?
Sure it would be more complex but I think it may be a small price to pay if it could provide an accurate unchanging bias current over all operating temperatures.
I think a uC is overkill unless you were going design a sustained Platue bias system like the krell amps.
But here are my comments anyway,
You mentioned you'd get the uC to measure the gate voltages. I think it would make more sense to measure the voltage drop across the source resistors. That way there would be no calibration you just pre-program the uC with the voltage you want there.
Re: Lateral V. Hex mosfets

HarryHaller said:

Seems that you answered your own question. At the risk of starting another riot, repeat after... me phase margin, phase margin, phase margin...... Good phase margin makes for stable amplifiers.


(Global) phase margin is always good but I doubt it is going to solve the problem Hugh describes. MOSFETs and fast, high-power bipolars can develop local oscillations if driven from a low impedance source.

A detailed analysis can be found in the 2000 edition of Tietze/Schenk: "Halbleiter-Schaltungstechnik". Unfortunately, the older editions, some of which were translated, do not contain this analysis in that much detail. You can have this effect even in the absence of any global feedback. In the end, it is still about phase margin, but only within a local, unintended loop.

I have experienced similar problems in my power amp that uses high speed Sanken bipolars as emitter followers. Under severe load conditions (i.e. 1 uF film from output to ground), the faster side (I think the PNP) would sometimes start to oscillate at >100 MHz at low amplitude, while the overall loop maintained stability.

One way of quenching the local oscillation was to increase base resistors, but turned out to slow down the output stage so that I would have had to roll off the voltage gain stage earlier to preserve phase margin under all load conditions. What worked perfectly well for me was to put a 2.5 mm ferrite bead on the base lead of each PNP output transistor. The amp is now stable with any kind of load (using only minimal inductive isolation on the output, three turns of 15 mm diameter) and I could keep the dominant pole at 25 kHz and the unity gain bandwidth at about 50 MHz.

Asking for trouble...

Well, I could have rolled the thing off earlier but then I wanted to obtain the highest bandwidth I possibly could. Having used lots of degeneration in the gain stages, getting the thing to be stable with a purely resistive input impedance and s 5 or 10 µH output insulation coil was pretty easy. But I wanted to preserve treble damping factor (still am not sure whether that is needed), so I set myself the goal for the amp to be stable with all kinds of loads and only a 0.25 µH coil. That was quite an effort, took me weeks to tweak the degeneration and the lead and lag compensation. The bode plot must look quite weird but what counts at the end of the day is the intercept angle at unity gain (there is a nice BB app note on that).

As a matter of fact, the local oscillation was quite helpful in adjusting the big loop. If it didn't break into a global oscillation after several seconds, the main loop was pretty stable. Another thing that helped in the end, all to my surprise was to use 10R base resistors between the input low pass filter and the diff pair input transistors, out of all places! Will post the circuit some day, once I succeed to find it...

Re: Asking for trouble...

I've had to do this for a number of complimentary symmetrical power amp diff input stages I've built over the years. I know somebody will cringe, but at this point, I just throw in 10-100 ohms for any such input stage I build and call it done :)

Curious: are you using JFETs or bipolars? What currents are you operating at and how much degeneration are you using? I encountered this more often with bipolar inputs and high amounts of degeneration.

capslock said:
....Another thing that helped in the end, all to my surprise was to use 10R base resistors between the input low pass filter and the diff pair input transistors....

uh oh, my reply was eaten by the computer crash (who ever pretended NT was stable???)!

I use SSM2210/2220 dual transistors (relaxed MAT02/03 in plastic case) at about 4 mA current, emitter degeneration is 27 R.

How come a complementary diff stage would be more susceptible than a single diff stage? For small AC signals they are very similar, after all.

Have you been able to figure out the differences between the Stoichino design and more conventional ones?