Hi John,
I have done a lot of simulations with MicroCap and found that the optimal voltage across the emitter resistor depends on the value of the resistor. In the old days, most people used a relatively high value, say around 0.47 ohms. This requires a bit higher voltage, something along the line of 20 mV to 25 mV for best results.
As the value of the emitter resistor is reduced, the optimal voltage across it drops somewhat. Nowadays, many amps use emitter resistors down as low as 0.1 ohms. The optimal voltage across this is in the range of 12 mV or 15 mV. Of course, with the lower resistor value it becomes much more important to match the output devices (assuming there are more than one pair!). Also the amp becomes more susceptible to damage if the output is shorted.
If you do the math, you will see that using a lower value of emitter resistor yields a higher idle current and more power consumption and more heat.
As always, everything is a tradeoff.
I have done a lot of simulations with MicroCap and found that the optimal voltage across the emitter resistor depends on the value of the resistor. In the old days, most people used a relatively high value, say around 0.47 ohms. This requires a bit higher voltage, something along the line of 20 mV to 25 mV for best results.
As the value of the emitter resistor is reduced, the optimal voltage across it drops somewhat. Nowadays, many amps use emitter resistors down as low as 0.1 ohms. The optimal voltage across this is in the range of 12 mV or 15 mV. Of course, with the lower resistor value it becomes much more important to match the output devices (assuming there are more than one pair!). Also the amp becomes more susceptible to damage if the output is shorted.
If you do the math, you will see that using a lower value of emitter resistor yields a higher idle current and more power consumption and more heat.
As always, everything is a tradeoff.
john curl said:
What I wrote is exactly Oliver's paper development and conclusions.
There is no optimal resistor or voltage drop but an optimal product of gm times R.
Of course if the base voltage is 0 then gmR=gmRe= IoRe/Vt.
IoRe is your voltage drop and we agree.
R=Re + Rb/(beta+1). You are wright that the nonlinearity in beta and the base voltage drop can change this optimum.
I know that all this is well known but it is interesting (IMHO) for some peole to understand why.
The main question of my post is:
The optimum being gmRe=1 this means that gm must stay constant with temperature and not Io ( bias) to minimise distortion in operation. Do you have experience with this.
JPV
JPV
Charles Hansen said:
I fully agree. That is exactly the conclusion in other words:
with Re 0.47, (R=Re + Rb/beta+1), R is about Re
then Regm=1 means ReIo=Vt=26mV at room temp
With Re droping the voltage drop on Rb/beta+1 is not negligeable compared to the voltage drop on Re. At optimum, the two must still add to 25mV therefore the voltage drop on Re must decrease.
We agree.
Still the question is: what of this optimum when temperature increases
JPV
JPV said:
Here is a related way to look at why the optimum RE voltage decreases as RE gets small. Take a look at the asymptotic transconductance of the bipolar transistor as Ic increases. For a typical RET, this maximum gm is on the order of 10 S. This is due to both Rb/(beta+1) AND to any internal emitter resistance, be it intentional or not. The inverse of this asymtotic gm can be thought of as if it were a small internal emitter resistor - as if it were a part of the external RE. Therefore, if the external RE is 0.1 and the internal BJT RE is 0.1, then the effective emitter resistance for which bias should be optimized is 0.2 ohms. We then place 26 mV across this 0.2 ohms, but realize that, in this case, only half of it appears on the outside, because only half of the RE is the external 0.1 ohms. So we end up putting 13 mV across the external RE.
Cheers,
Bob
Charles Hansen said:
Hi Charles,
I’m sorry to hear that you’ve had some difficulty with vertical MOSFETs. I really appreciate your candor in discussing some of your experiences, as this is the way many of us on this board learn going forward.
I think we have to agree to disagree on some of your points. In particular, your characterizing a whole technology, like vertical MOSFETs, as designed for switching applications and by implication therefore not good for audio applications is something that I don’t buy at all.
Within every technology, some parts are optimized for one application over another, and this is just as true for biploars as well as vertical MOSFETs. This optimization may take the form of what aspects of performance are specified and characterized in more detail, or it may take the form of engineering tradeoffs in the design of a particular device, such as the target Rds-on for a vertical MOSFET. While this grain of truth exists for some vertical MOSFETs, it seems inappropriate to make generalizations about the technology based on this.
It is certainly true that vertical MOSFETs’ main application technology is in switching applications, where their natural speed and freedom from minority carrier storage makes them excel over bipolars. That is also the biggest market for them, so many of the parameters specified on the data sheets are less relevant for audio and more relevant to switching. It is also true that as the vertical MOSFET technology has progressed, there has been a trend to greater availability of devices best suited for switching. One example is the TrenchFET, which I have previously argued is not well-suited to audio applications.
None of this argues that the IRFP240 and similar vertical MOSFETs are not fine devices for audio. It’s all about picking the right device and applying it optimally.
In the case of the IRCP054 device you used, this is a vertical MOSFET that is plainly less suitable for audio than the IRFP240 type devices because of its emphasis on an inordinately small Rds-on. I’m not sure why you chose this device.
There is a common misconception among some audio designers that the way to get high transconductance and low output impedance from a vertical MOSFET source follower is to pick a device with a very low Rds_on. In fact, Rds_on is a specification that is more relevant to switching applications and of very limited relevance to audio applications. Rds_on only sets a limit on the asymptotic maximum transconductance available from the device at very high currents. Below this limit, where it matters most for linear applications, transconductance of a vertical MOSFET is mainly set by its operating current. Moreover, trying to get too much transconductance out of a single device can result in serious thermal stability problems.
A transistor in a typical output stage forms a thermal feedback loop that has a positive feedback factor (at least where it has a negative temperature coefficient of control voltage for a given operating current). This is true for both BJTs and MOSFETs. This can easily be seen as follows: Assume that a MOSFET is at a given idle bias current and temperature. Then assume that the junction temperature increases by 1C. That will cause an effective gate voltage change, typically by 5 mV/C for a vertical MOSFET operating at typical bias currents. That will in turn cause an increase in drain current in accordance with the transconductance of the device. The current increase will cause an increase in power dissipation in accordance with the Vds across the device. Finally, the increase in power dissipation will cause an increase in junction temperature in accordance with the thermal resistance from junction to heat sink.
If this last temperature increase is 0.5C, for example, we have a positive feedback Beta factor of 0.5. If this last temperature increase is 1C, we have thermal runaway. This feedback factor should never be allowed to be greater than 0.5. BTW, for this simplified analysis of thermal loop gain, we assume that the heat sink temperature does not move, and we assume that the output node does not move, it being relatively held in place by the low output impedance of all other paralleled devices. In this respect, this is a bit of a worst case, but it seems appropriate. The Beta factor can be simply expressed as:
Beta = Theta_JS * Vds * TCvgs * gm
This is an inverse measure of the intrinsic thermal stability of the device at its operating point in its local thermal and electrical environment.
For a T0-3P 150 W device with an average insulator, Theta_JS may be on the order of 1.3C/W. For a MOSFET operating at 50 V at an idle bias of 150 mA and a gm = 0.6 S, we get a Beta of 0.2, a relatively safe value.
On the other hand, suppose we are operating at 400 mA with a gm of 1.6 S. We now have a Beta of 0.52, and we may be in trouble.
For a bipolar operating with a 0.1 ohm RE and a net gm (including RE) of 5.0 S, with Vce = 50V, and a TCvbe of 2.2 mV/C, we get a Beta of 0.72, a very unsafe value.
For both BJT and MOSFET, the value of transconductance may increase at points away from idle bias, increasing Beta and increasing the danger. The bottom line here is that if we try to get too much transconductance out of a single device, we may be asking for trouble.
You mentioned thermal instability problems with the IRCP054 device, specifically citing rising drain current with Vds above 30V, even though the device is rated at 60V. I can’t explain that phenomenon (unless your curve tracer gave the device junction a few ms to heat up during the sweep). But I also don’t understand why you were using only a 60V device in a power amplifier. What was the reason you chose that device?
Nevertheless, the rise you described is odd, and I have not seen evidence of it in IRF240 type devices. The Fairchild datasheet for the IRFP240 shows drain current flat out to 100V (why they don’t show it to 200V for a 200V part I don’t know). I agree that it is unfortunate that IR shows output curves up to only 30 V for the IRCP054 device. As a result the presumed presence of the rise on the spec sheet would go unknown to you. That is a huge device, with an Rds_on of only 0.014 ohms and an input capacitance of 7000 pF at low Vds. I would agree that that particular device is certainly not optimal for audio. Its transfer characteristic on the spec sheet doesn’t even go below 4 A.
Anyway, I think the problem you had here was more related to individual device choice and their application than to vertical MOSFETs being less suitable for audio. BTW, from the very beginning, Siliconix put out application notes touting vertical MOSFETs for audio. Just ask Ed Oxner.
Cheers,
Bob
Yeah, Chas, why did you use a 60V device? Could it be that you had something like +/- 25V power supplies and low voltage mosfets have higher transconductance? I could be wrong, but that is why I used 100V mosfets with a +/- power supply of 45 volts. Perhaps, if I used 200V devices that Bob seems to think as a minimum requirement, I would have had more success.
.
Actually, I'd like to see some comparative transconductance data at lower currents. I've used IRF devices in tube circuits with some pretty good results, but I have wondered if the transconductance advantages of device A versus device B hold up at currents below full pulse on, where they're usually spec'd.
SY said:
To get transconductance off of the spec sheets, one usually has to manually calculate the slope of the Id vs Vgs transfer characteristic, such as from Figure 7 in the Fairchild IRFP240 data sheet. They are kind enough to show gm vs Id in Figure 12, but it is difficult to get gm off of there at low currents in the range of 100 mA.
I have done this for numerous vertical MOSFETs and the gm is fairly predictable when viewed as a function of current. We know that for a BJT, gm = Ic/26, where Ic is in mA. I have found that for vertical MOSFETs, gm = IC/k, where k typically ranges from 250 at low currents to 500-1000 at medium-high currents. In other words, gm for a vertical MOSFET is about 1/10 that of a bipolar at low currents (less than 200 mA, for example). These are only approximations, of course, and the fact that k is a function of current must be kept in mind.
(the value of k increases less with Ic at high currents if you first take out a value of effective resistance equal to a large fraction of Rds_on).
It is NOT generally true that at the same low current (say, 100 mA) a device with a low Rds_on has a higher transconductance. It is closer to the truth to say that they will both have about the same transconductance at these low currents.
Cheers,
Bob
Here is another data point from some old notes of mine for lateral MOSFET's:
~~~~~~~~~~
Transconductance of Lateral MOSFETs (Magnatec BUZ903) as measured on Tek 576:
1 mA => 21 mS
2 mA => 35 mS
5 mA => 70 mS
10 mA => 110 mS
20 mA => 160 mS
50 mA => 270 mS
100 mA => 380 mS
200 mA => 550 mS
500 mA => 800 mS
Square Law is pretty good from 10 mA and up. Below that the transconductance drops.
Transconductance divided by square root of current = ~1.1 for the lateral MOSFET.
~~~~~~~~~~
I had done some similar measurements on some vertical parts, but they were just written on a whiteboard and it was erased several years ago. I seem to recall that we were looking at transconductances around 5x lower than a bipolar at the same current. This is relatively close to Bob's figure of 10x lower. (Kind of like what astronomers tell each other, "What's a few orders of magnitude between friends?")
~~~~~~~~~~
Transconductance of Lateral MOSFETs (Magnatec BUZ903) as measured on Tek 576:
1 mA => 21 mS
2 mA => 35 mS
5 mA => 70 mS
10 mA => 110 mS
20 mA => 160 mS
50 mA => 270 mS
100 mA => 380 mS
200 mA => 550 mS
500 mA => 800 mS
Square Law is pretty good from 10 mA and up. Below that the transconductance drops.
Transconductance divided by square root of current = ~1.1 for the lateral MOSFET.
~~~~~~~~~~
I had done some similar measurements on some vertical parts, but they were just written on a whiteboard and it was erased several years ago. I seem to recall that we were looking at transconductances around 5x lower than a bipolar at the same current. This is relatively close to Bob's figure of 10x lower. (Kind of like what astronomers tell each other, "What's a few orders of magnitude between friends?")
Bob Cordell said:
Yes, but the ONLY vertical parts that I know of that were specifically designed for audio were the Toshiba K1530/J201 pair.
Bob Cordell said:
I wanted a balanced bridge output stage with only four transistors. I didn't want to have to match a bunch of parts. So I found the biggest, baddest, beefiest P-channel part available at that time, a Harris part rated at 60 amps at 60 volts. (I only needed 40 or 45 volts to get 100 watts with a balanced bridged output stage.)
Since there weren't any audio parts, I had to find my own complement. I ordered dozens of likely sample candidate N-channel parts and looked for the best match (transconductance-wise) with the Harris by looking at them on the curve tracer. The IR was the best match, plus you could get it in two versions -- one regular and one with the Sense-FET thingy. I thought it would be cool to use the Sense-FET to check the bias, as then I could omit the source resistor altogether.
Worked pretty well, all things considered.
Bob Cordell said:
One thing I learned early on -- NEVER, EVER trust the manufacturer's curves. They might give a rough guideline at best. The only way to know how it performs is to throw it on your own curve tracer. In this case, the phenomenon was real and not due to heating. As I noted in a previous post, after looking at dozens of potential substitutes, I found one that had VERY flat characteristic curves (although still not quite as flat as the P-channel Harris parts were -- for some reason P-channel vertical MOSFET's suffer less from this problem than do N-channel parts.)
john curl said:
Yes, exactly. 25 volt rails easily produced 100 watts with a bridged output stage. And for a given die size, a lower voltage part will handle higher currents. By using low-voltage parts, I could get away without having to parallel devices and that meant no matching was required. A nice labor-saving idea, especially since FET's (both J- and MOS-) are all over the map.
SY, I would not trust any predictions below 10mA on devices rated over 1A. Vertical devices should be worse in this regard as they approach the ideal transconductanse that bipolars have. The curvature will become more bipolar like, as you cannot exceed the ideal Gm for a given current. At least, not to my knowledge.
Charles Hansen said:
Thanks, Charles.
Although I haven't looked hard at Lateral MOSFET gm, it has been my suspicion that the gm, when viewed as a function of current, and when the resistive component is properly taken out, is pretty much the same as that for a vertical. Your numbers seem to confirm that. Your number at 100 mA appears to be almost dead nuts on to 1/10 the gm of a bipolar at the same current, and is very close to what I see for a vertical, for example.
Thanks,
Bob