I am looking for a P channel mosfet for use in a class A or AB audio amp application. Quiescent Vds can be up to 100 volts with peak excursions approaching 200 volts at near zero current. Quiescent current is at or below 160 mA in class A with peaks to over 300 mA near saturation (near zero Vds). Class AB applications will see similar or lower static bias currents with peaks approaching 1 Amp, again at Vds in the 10 volt range.
The application is a vacuum tube amp with a cascode output stage involving a tube paired with a mosfet. The schematic is in post #91 here, the fet in question is the IRF9640:
https://www.diyaudio.com/forums/tubelab/340856-unset-coming-10.html
ON Semi / Fairchild does NOT seem to acknowledge secondary breakdown in their mosfet data sheets, which is real. They no longer list any mosfets for high powered audio amps on their web site. Many other companies do not show any DC curve in their SOA specs. I have tried dozens of the usual mosfets intended for SMPS use, and I have a big pile of dead ones. The Fairchild FQP9P25 is the best I have tried so far, but I have scattered a couple by driving the snot out of the amp with a guitar preamp, or having a bad ground on an interconnect and messing with it while playing music at full power. Yes, these are extreme tests, but I build my stuff to be capable of this kind of abuse.
I spent a good deal of the last year of my engineering career (2013) chasing secondary breakdown in GaN fets used in RF amps......now I'm chasing it again, or running from it.
The application is a vacuum tube amp with a cascode output stage involving a tube paired with a mosfet. The schematic is in post #91 here, the fet in question is the IRF9640:
https://www.diyaudio.com/forums/tubelab/340856-unset-coming-10.html
ON Semi / Fairchild does NOT seem to acknowledge secondary breakdown in their mosfet data sheets, which is real. They no longer list any mosfets for high powered audio amps on their web site. Many other companies do not show any DC curve in their SOA specs. I have tried dozens of the usual mosfets intended for SMPS use, and I have a big pile of dead ones. The Fairchild FQP9P25 is the best I have tried so far, but I have scattered a couple by driving the snot out of the amp with a guitar preamp, or having a bad ground on an interconnect and messing with it while playing music at full power. Yes, these are extreme tests, but I build my stuff to be capable of this kind of abuse.
I spent a good deal of the last year of my engineering career (2013) chasing secondary breakdown in GaN fets used in RF amps......now I'm chasing it again, or running from it.
Copy what the high end crowd did in 1974 with rather cruddy BJTs: stack transistors in series to cut down the voltage stress; connect several of those stacks in parallel to reduce peak current, peak power, and SOA stress. The most famous example is James Bongiorno's "Ampzilla". Here is another example.
Click on the image to see it full size and undistorted.
_
Click on the image to see it full size and undistorted.
_
Attachments
Last edited:
Still not ready to give up and use a combination Mosfet/bipolar (CFP or darlington)? The MJL4281 will hold up to the voltages and currents you need. Been using them for solid state amps which run off *tube voltages* and drive low impedance speakers. Mostly arc welders/ woofer cone launchers/ voice coil cookers where a little crossover distortion isn’t the end of the world.
I’d be curious as to whether they are good enough for your unset driver, sound quality wise. Bipolars do have a nasty region of non linear capacitance when they cut off, but if that is somehow prevented….
I’d be curious as to whether they are good enough for your unset driver, sound quality wise. Bipolars do have a nasty region of non linear capacitance when they cut off, but if that is somehow prevented….
Using a bunch of 60 or 80 volt transistors, which was all you could get back then, except unreliable switching types not available in PNP at all. These amps suffered from dynamic balance issues - at high frequency it didn’t divide evenly anymore and BANG. Especially if it broke into oscillation like the Tigersaurus did. I’ve done this successfully to extend SOA and allow +/-125 volt operation with MJ15024’s, but an individual device will actually block 250 volts for a few us at a time, so square wave drive won’t blow the amp. Those old 80 volt epi-base devices won’t take it. MJ4281’s will, and the fT is pretty high.
Of course not new in 1974. 11 years before that, Burwen published 64 devices (all PNP!) series/parallel for 1,000 Watts. He did not take series real far: bridged on +/-15V. That made impedance awful low but his app allowed an output transformer (even so, lead-inductance ate his sizzle). All this in a rack the height of a man. (What would it be today, 1U?)
Attachments
IRFP9240 is P channel mosfet.
Laterals suffer less from secondary breakdown.
Lots at profusion.com
Laterals suffer less from secondary breakdown.
Lots at profusion.com
There is a good reason several commercial designers abandoned vertical mosfet outputs and switched back to more expensive bipolar.
I made a monster solid state amp in about 1971 when I found a humongous heat sink in some military scrap with 24 X 2n3773's on it with a fan on each end. They were wired in 4 parallel banks, each with it's own emitter resistor. I simply wired them as two banks of 12 parallel transistors and used the transformer driven totem pole circuit often seen in 10 to 25 WPC germanium transistor amps and receiver's of the late 60's. The power supply was something that can not be discussed here, but it consisted only of a BIG bridge rectifier and some REALLY big caps. As expected the power output depended on how many speaker cabinets were connected up. It was likely in the kilowatt range into a 2 ohm load.
We drove the beast with a 50 watt Marshall tube guitar amp and hooked up four 8 ohm cabinets wired in parallel with 4 X 12 inch guitar speakers in each wired for 8 ohms. It filled a large school's outdoor football stadium full of 60's rock music loud enough for the cops to order it to be turned down. The band creating the racket for a local city's 4th of July celebration would use that set up for their PA for several years.
In today's case I need a better part for an existing PC board design, which is laid out for a single P-channel TO-220 mosfet. The board works fine unless I put the biggest and baddest sweep tubes ever made in it and push the supply voltage into the 550 to 650 volt range where I get 30+ watts from a single tube in SE, and 250+ watts from a pair in push pull.
I trying to build a kilowatt (500 WPC) tube amp before I get too old to be able to move it. A new PC board design will be used, and some of these tricks will be tried, but the published SOA data on many PNP BJT's doesn't look so good either. I do have some MJW21195 around here somewhere.
Mosfet powered amplifiers running from +/- nearly 100 volt rails exist, I assume that someone here knows what kind of parts can be used to fix them today.
IRFP240/9240 were probably the best hexfet types for linear amplifiers. They claimed not to have any second breakdown. Their real SOA was similar to many of the bipolars that were used at the time - like the really good Toshiba and Sanken parts. Not free of second breakdown, but high enough for it to be useable. Parts made for switching applications today can’t even do this, and tube applications quite frequently need something good for more than 200 volts. The On Semi 4281/4302 have the best SOA between 100 and 200 volts of any bipolar out there, and boasts a 350 volt VCEO. Won’t handle full power up there by a long shot, but will take more than a hexfet will - even an IRFP240. 21195/6 are better between 80 and 100 volts, but not as good up at 200.
Lateral MOSFETs are immune from secondary breakdown, perhaps an ECX10P20? Only rated to 200V though. Datasheet has SOA plot.
ECX10P20 Plastic Lateral MOSFET | Profusion
datasheet:
http://www.exicon.info/PDFs/ecx10p20.pdf
ECX10P20 Plastic Lateral MOSFET | Profusion
datasheet:
http://www.exicon.info/PDFs/ecx10p20.pdf
That schematic lacks a zener to protect the gate-oxide - probably why you have a pile of dead MOSFETs.
Last edited:
The schematic is from the LT spice simulation where parts never blow up. The real amp does indeed have a 15 volt zener from gate to source.
New experiments tonight may have revealed the cause of some or all of the blown fets. I managed to stuff enough current limiting resistors in various places in the circuit to get it to live when it goes bonkers. A strong (into clipping) oscillation in the 35 to 45 KHz range causes the tube current to jump into the nearly 1 Amp range. Fixing an open ground on the OPT secondary killed the oscillation and allowed me to abuse it seriously with the power supply set at 520 volts without issue.
I ordered some of those Exicon fets. They might be a good choice in the big amp.
The double die Exicons are probably worth considering, too.
Multiple pairs at +/-90VDC are likely workable.
Multiple pairs at +/-90VDC are likely workable.
That’s rather pretty - full power to 90 volts. Way better than the IR stuff. You do pay for it though - Rds(on) about half an ohm instead of the usual 50 milliohms.
When I’m digging for mosfets that can safely be use as screen regulators, first thing I look for is particularly high Rds(on) values.
When I’m digging for mosfets that can safely be use as screen regulators, first thing I look for is particularly high Rds(on) values.
Hello Tubelab_com. Impressive endeavor and power amp to be. Can you reconfigure to use N-MOS instead; which will open the door further for the reputed robust SiC N-JFETs?
The semiconductor is a follower in the cathode circuit of a tube. A P channel fet or a PNP BJT are the easy choices here. Any N type device would be used in a common source or emitter configuration here it's active characteristics play a large role in the overall circuit's characteristics. Experiments along this route several years ago failed to produce the desired overall triode like curves from the composite pair.
The 20P50 device looks good except for it's pretty large Crss, especially at low voltages. It's little brother, the 10P50 has more manageable Crss, and better SOA ratings in the area of interest.
I ordered some of the Exicon fets, and I will get some 10P50's. Digikey does not have any in TO220, haven't checked Mouser yet, but the Exicons are TO247, so these will be for the modular version of the UNSET which targets BIGGER power levels.
After sending several mosfets to their death over the last 4 days I have traced the issue to a 40 KHz oscillation at full power. The big Hammond 1628SEA is a near dead short at that frequency which sends the tube current into the hunndreds of mA range. Now that it's fixed I have not been able to blow any mosfets in a 20 Watt SE amp on 520 volts. I have not gone higher yet due to the 500 volt filter caps in the board.
So it was premature to blame the Pchannel MOSFETs and unnecessary to search for different parts?
Not really. The P-channel fets that I have been using are designed for SMPS use. They have been largely reliable at audio output power levels at or below 20 watts in a class A SE amp and 100 watts in a class AB push pull amp. I plan to build bigger amps and would like to use a part that was actually intended for continuous operation in the linear region.
My experience with GaN RF fets designed for pulse radar at 100 watts in continuous operation at 8 watts resulted in a two year project between us (Motorola) and the component maker which resulted in a new part design.
I will test the IXYS and Exicon parts, and any other suggestions that look like a better fit than what I currently have.
I looked up the TO-220 version of the IXYS part and it is nearly a year in backorder. It is doubtful that my current driver design can feed the Crss of them any way, and the Crss of the Exicon part are not specified at low Vds voltages.
My experience with GaN RF fets designed for pulse radar at 100 watts in continuous operation at 8 watts resulted in a two year project between us (Motorola) and the component maker which resulted in a new part design.
I will test the IXYS and Exicon parts, and any other suggestions that look like a better fit than what I currently have.
I looked up the TO-220 version of the IXYS part and it is nearly a year in backorder. It is doubtful that my current driver design can feed the Crss of them any way, and the Crss of the Exicon part are not specified at low Vds voltages.