traderbam said:That's one way to settle the argument. Very witty. Perhaps you'd better throw a tube in there just in case.
You are right. Tube drives this device. 12L6 in triode connection loaded by transistor CCS.
Speaking of testing, it's pitty that I did not take a picture of evaporized tip of scredriver and binding posts. Also, fuses throw glass peaces inside of the amp.
Bob Cordell said:
I certainly agree there. I just wanted to point out a caveat of using the NMOS and PMOS models for vertical MOSFETs.
In general, there's some really bad models floating around. The MJL3281A and MJL1302A models from ON Semiconductor have a number of problems that I've documented here. For the MJL3281A model, the simulated fT is low by a factor of six, and the beta vs current is all wrong.
In another case, the Fairchild models for the KSA1220A and KSC2690A have TF not specified, making it default to zero. This gives simulated fT values of about 1 GHz at midrange currents. A quick way to see fT in LTSpice is to do an operating point sim, then View, SPICE error log. There's lots of other handy info in that log for an operating point sim as well.
Bob, I normally use fets everywhere, except at the very output. I have also designed power amps that are 100% fet, with Vfets, vertical mosfets, and lateral mosfets as output stages for different designs. I did this between 20 and 30 years ago.
This is what I found:
Vfets are now unavailable, but interesting devices.
Lateral mosfets are rugged, fairly linear, but have too low of transconductance to be optimum.
Vertical mosfets have much promise but they won't hold together when an amp is accidently shorted, at voltages over about +/- 35V.
In fact, most manufacturers do not recommend vertical power mosfets for linear operation these days. I wonder why?
IR makes a lousy power p channel mosfet.
The breakdown mode in mosfets occurs faster than second breakdown in a well designed transistor output amplifier, and requires at least as much protection as a transistor output amplfier.
I have been using an output stage that is a combination of vertical power fets as drivers and multiple pairs of ring emitter transistors. I use 9 pairs of output transistors in the JC-1 power amp. I can produce more than 800W into 4 ohms.
My slew rate, without an output coil, is more than 100V/us
I have made a ring emitter based power amp in the past that has single sided slew rate of greater than 500V/ us. I used a 2uH inductor to achieve this slew rate and still have it stable.
I run +/- 90V on my 8 amp driver vertical mosfets. Can anyone show me how these devices can be connected directly to the output, without blowing up the first instant someone accidently shorts the amp?
Inquiring minds what to know!
This is what I found:
Vfets are now unavailable, but interesting devices.
Lateral mosfets are rugged, fairly linear, but have too low of transconductance to be optimum.
Vertical mosfets have much promise but they won't hold together when an amp is accidently shorted, at voltages over about +/- 35V.
In fact, most manufacturers do not recommend vertical power mosfets for linear operation these days. I wonder why?
IR makes a lousy power p channel mosfet.
The breakdown mode in mosfets occurs faster than second breakdown in a well designed transistor output amplifier, and requires at least as much protection as a transistor output amplfier.
I have been using an output stage that is a combination of vertical power fets as drivers and multiple pairs of ring emitter transistors. I use 9 pairs of output transistors in the JC-1 power amp. I can produce more than 800W into 4 ohms.
My slew rate, without an output coil, is more than 100V/us
I have made a ring emitter based power amp in the past that has single sided slew rate of greater than 500V/ us. I used a 2uH inductor to achieve this slew rate and still have it stable.
I run +/- 90V on my 8 amp driver vertical mosfets. Can anyone show me how these devices can be connected directly to the output, without blowing up the first instant someone accidently shorts the amp?
Inquiring minds what to know!
traderbam said:
Brian,
Sorry if I mis-interpreted what you said.
I guess I shouldn't be surprized that my mention of differences between MOSFETs and bipolars sparked a big debate, but I'm frankly surprized that my assertion that MOSFETs are faster than bipolars was so controversial. Help me to better understand your point, and let me clarify my position and try to better understand yours.
If I read you right, you agree that ft is an appropriate measure of speed for a MOSFET. I think you also agree that the ft of a MOSFET is quite a bit higher than that of a ring emitter bipolar (on the order of 300 MHz vs 30 MHz under a reasonable set of operating conditions, give or take a factor of two).
What it sounds like you are disagreeing with is the assertion that ft is an appropriate measure of the "speed" of an output device for an audio power amplifier. Am I correct so far?
I need to better understand what your preferred metric for relevant speed is.
I believe that, regardless of which device type we are talking about, we need to look at small-signal bandwidth of the output stage in its circuit environment, and we need to look at the large-signal turn-off rate of the device in its circuit environment. These two may be different, but they are both important. Although they are different, they are usually related, since device capacitances influence both of them.
In the small-signal simulation I suggested, the 50 ohm resistor was not put there for stability reasons, but rather to assess the transfer bandwith of the stage under a reasonable set of conditions. Driving both devices with a voltage source, for example, would give a meaningless result. There is no VAS-driver combination that I know of that has a zero output impedance. Having the 50 ohm resistor there allows the a.c. current gain of the device to properly come into play in the analysis. I'm OK with it if you disagree with this approach to asessment of the small-signal speed, but tell me what you propose as an alternative and why, so I'll be able to understand and evaluate your point.
For large-signal speed, what I care about most is being able to turn the device off fast enough, especially in the crossover region. For bipolars, this means sweeping out the minority carrier charge which is incrementally represented by the hybrid-pi base-emitter capacitance, which is in turn determined by the ft of the device. For a bipolar, the available turn-off current is usually set by the standing current of the driver, and will typically be between 10 mA and 50 mA (assuming no cross-coupling capacitor tricks, for now, to keep things simple).
So to me, the amount of reverse base current needed to take the device from one amp down to 0.1 amp is a good metric. This is basically how long it takes the reverse base current to discharge the average base-emitter capacitance, over this current range, by 60 mV. For a 30 Mhz ft bipolar starting at 1A, the Cbe is 0.2 uF and ends up at 0.02 uF when we reach 0.1A. A working average of 0.1 uF is probably reasonable for back of the envelope stuff. With a 10 mA reverse base current, we end up with about 600 ns.
For a MOSFET with the same available turn-off current, it is the time it takes to move the gate by about 700 mV against the Cgs and Cgd capacitances. If those capacitances add up to about 1300 pf average over that 700 mV Vgs range, then we get about 90 ns. That is why I think MOSFETs are faster. If I've screwed up here, please let me know how. It certainly wouldn't be the first time I screwed up a back-of-the-envelope calculation.
One thing that you said I would like you to explain better. You said, "for a given ft the speed of a bipolar is related to gm and beta. So you miss a trick if you compare only on ft." What is the trick we are missing? It sounds like you are somehow concerned about f-beta affecting speed for a given ft. This leaves only beta affecting speed, since f-beta ~ ft/beta. But beta has very little influence on how long it takes to pull the minority carrier charge out of the base. Are you perhaps suggesting that you want lower beta for higher speed, like we did in the old days when we used a little bit of gold doping in bipolar digital ICs to kill the minority carrier lifetime? Help me out here.
Finally, I completely agree that, in the absence of error correction, you have to run MOSFET output stages significantly hotter than what you can get away with for bipolars. That is a result of the big drawback of MOSFETs called transconductance droop. That is probably the single biggest reason I pursued the error correction scheme. In fairness, however, recognize that there are many very high-quality bipolar amplifiers that run their output stages hot also, in Class A-AB, so they are in Class A up to perhaps a couple watts.
Cheers,
Bob
Folks, if you want to understand why power vertical mosfets tend to break with higher voltages, just compare the 10ms safe area of the best high speed bipolars to any other pair of practical vertical mosfets.
What I have found: 10ms-100V
2SK3264---5A
IRF130----1.7A
IRF140----2.8A
2SJ201-----3A
Please show me a better vertical power mos pair!
What I have found: 10ms-100V
2SK3264---5A
IRF130----1.7A
IRF140----2.8A
2SJ201-----3A
Please show me a better vertical power mos pair!
john curl said:
John,
Thanks for this detailed answer. I agree, when MOSFETs go, for whatever reason, they go fast. Hey, I've blown up a lot of them! My experience has been that they will usually not outlast a fuse; they definitely need some sort of last-resort protection against short circuits, such as an electronic crowbar. In some respects, in a short circuit condition, they may be their own worst enemy, since they will readily each attempt to deliver 50 amps or more; there is not much of a mitigating mechanism to reduce such high current flowing in the presence of a sudden short. I don't know how much of a role it mught play, but high-current beta degradation in bipolars might be an example of such a mitigating factor (as long as the driver transistor survives - in this sense, your choice of MOSFETs as drivers makes sense).
As far as device destruction goes, sometimes it is a chicken-egg situation, and it is hard to tell what the initial proximate cause of the destruction was. For example, often in conventional output stages the driver will die with the output transistor, and one doesn't always know which one went first. Another way in which MOSFETs will go very fast (fast devices blow faster?) is if for any reason the oxide is punctured. For example, in the case of a short circuit, if the driver successfully applies more than 20V forward bias to the MOSFET gate, poof!! That wasn't the fault of the MOSFET, but rather an ill-controlled driver. Similarly, if the application of the short circuit causes operating points or circuit parameters to shift enough to cause the MOSFET to go into a parasitic oscillation at 100 MHz, poof!! All I'm saying here is that we don't always know that the cause of the destruction was the die temperature getting up to beyond, say, 175C.
The p-channel devices are usually inferior to the n-channel devices, but what specifically do you especially not like about the IRF p-channel MOSFETs?
Here's a trivia question for you that I honestly don't know the answer to. Which came first, the HEXFET or the ring emitter transistor? Both were born of the same essential idea: use integrated circuit-like technology to effectively build a transistor that consists of hundreds or thousands or tens of thousands of little transistors in parallel.
I'm still interested in learning what your favorite difficult load is for evaluating your amplifiers, and what general type of output stage protection circuit you use.
Bob
mikeks said:
Mike,
There is nothing wrong with the skillset of most of the people here. I continue to be impressed by everyone's intelligence. And there is nothing wrong with intelligent, well-intentioned people disagreeing. Indeed, I have always asserted that what influences an EE's design the most is what he or she FEARS the most. This is certainly the case with me, and most of the engineers I work with agree with this observation.
So it is not so much skillset, or intelligence or even diligence. It has a lot to do with experience with a particular thing. We pick our poison. We find our comfort zone.
Anyway, can you elaborate on why you think that MOSFETs are inferior to modern BJTs in audio frequency applications?
We ALL learn from these lively discussions.
Thanks,
Bob
anatech said:
Chris,
I agree, there are a PLETHORA of really bad MOSFET designs out there. As is often the case with any "new" technology there are those numb-nuts who ignornatly exploit it because it is the latest buzz-word (digital, anyone?). There are also those ignoramuses (sp?) who see that the beta of a MOSFET is infinite and think that they therefore can make a good amplifier without using a driver stage between the VAS and the output MOSFETs.
Bipolar transistors suffered from these same sorts of things for years in the early days (indeed, some still do).
Bob
MikeB said:
And I think that should have been 2SC3264 in John's post (not 2SK).
There is substantial difference between MOSFET technologies. For instance, take IRFP240 (keep in mind this is in a smaller case compared to the bipolars mentioned, and specified 150W at 25C), we have at 100V / 10ms:
Original IR: 3A
Fairchild IRFP240B (Specced 180W @25C): 4.7A
Interestingly:
IR IRFP140: 3.1A
IR IRFP140N: 2A (it is specced 140W though).
It should also be noted that IR does not publich DC SOA for their parts at all (but Fairchild does). I'm not even going to begin comparing prices for 2SC3264 (which is no doubt an excellent BJT) and the lowly IRFP parts.
You could look at this in another way: make a single pair output stage. Take nearly any sensible device and you are limited to about +-40V rails calculating for a nominal 8 and actual 4 ohm load. Then, compare taking care of the price bracket, and you will find that it is very similar - in fact, although HEXFETs sometimes get a fraction of an amp ahead, in either case, you are well advised to provide some sort of proper SOA limiting. Basically, comparing IRFP240 and 2SC5200 and similar devices is a fair comparison.
That being said, I have never had a current limited MOSFET not survive a fuse (sensibly chosen, of course!), but I do take care of one thing: avoiding gate punchthrough either by oscillation of overvoltage. The failure rate of HEXFETs in designs where this is not assured, is in my experience 100%.
What is important is what works. Sometimes, a Japanese part will be expensive, but since I have my power amps made in Taiwan, the parts are what the Taiwanese select as what is easily available. Fairchild (IR) devices are superior, and what I normally use for audio applications.
I am happy that the Toshiba 2SJ201 parts are really better than I first thought. That leaves it easier for me to use them in future.
I would be careful however, I have been 'burned' before.
I am happy that the Toshiba 2SJ201 parts are really better than I first thought. That leaves it easier for me to use them in future.
I would be careful however, I have been 'burned' before.