Matching MOSFETs

[DaveR]

Power MOSFETs suitable for use in analog circuits have been hard to come by for some time now. As many of you know the amps I use and built use them as the output devices. I found some at a reasonable price from a reliable source and purchased some. Using multiple pairs in the output requires matching. I put together a little test fixture that allows me to take advantage of some test equipment that I have to match them quickly and easily.

The first picture shows the whole setup except the power supply. The digital function generator allows me to output short pulses at variable rates and adjust the output level in .001 Volt steps. The digital scope has the capability of taking a variety of measurements, the two I'm using are high and low.



The device currently under test is an N-channel so I'm pulsing the gate high, yellow, and reading the voltage on the drain, blue. The supply is set to 40 Volts, the load is a 10 Ohm resistor, so a 20 Volt negative pulse is two Amps, 15 Volts is 1.5 Amps and so on. In this picture I haven't dialed in the pulses to be exactly 15 volts.



And this is the test fixture itself, very simple. I diode clamp the drain to the supply and have a .1uF cap from the drain to ground to eliminate ringing. The switch with the black paddle switches between diodes pointing in opposite directions. The switch with the gray paddle turns voltage to the drain on or off. The white connector can be used to connect the fixture to the drain and source of the FET but I found it just as fast to use the clip leads.



A ten microsecond pulse 200 times a second at 2 Amps barely changes the temperature of the FET.

I have a spreadsheet that I record all the results on and helps identify the best matched groups. If you have any interest in seeing it let me know and I'll email it to you.

[FerdinandII]

Since the MOSFETS will be heating up in service, do you also need to match the TEMPCO?
What temperature would they typically idle at?

[DaveR]

At a half Amp bias they're typically running at +30 degrees F over ambient. This is installed and with music playing, not loud. The amps are fan cooled, although you can't hear the fan from the front of the amp. I expect the temperature coefficient to track pretty closely within matched groups. My ex-partner says these have a negative temperature coefficient so I would expect them to be self balancing.

[Roscoe Primrose]

My ex-partner says these have a negative temperature coefficient so I would expect them to be self balancing.

MOSFETs generally do have a negative temp. coef. which is a huge benefit in applications where high reliability is important...

[HAL]

If someone needs to match the N and P versions of Hitachi 2SK135/2SJ50 family MOSFETs, I have a Hafler Matching Box to grade them.

I tested and matched spares for my Hafler DH200/DH220 and XL280's amps, but works for all the other amps using the TO-3 case.

The grading number is just the switch position on the box when you test for closest to 100mA bias current.

[dberning]

NO NO NO!

See https://www.onsemi.com/pub/Collateral/AND8199-D.PDF
While this reference does not specifically discuss multiple parallel devices as much as hot-swapping, the principles apply.

The temperature coefficient of MOSFETs has a crossover point whereby parallel devices have enhanced current balancing and on the other side of this point a degraded sharing capability. If you are doing switching, you gain by this, but if you are operating in the active region you have the same problems with current sharing as bipolar transistors as the threshold voltage drops as the temperature rises, just like the base-emitter voltage drops as the temperature rises in a bipolar transistor. At low currents (relative to the current rating of the device) and high voltage (voltage higher than that obtained under fully switched on conditions for given load) you are in this active (analog) region. This means that the hottest device of a matched set will conduct more current and this tends to cause runaway. Same thing happens with bipolar transistors. Soooo emitter or source ballasting resistors are required for analog applications to suppress this effect.

Interestingly, this is why the modern MOSFETs are poor choices for audio class A, AB, and B circuits, but good for Class D. The newest MOSFETs are very fragile in their active regions but robust as switches. Also, in part because the transconductance is so high, they have poor linearity. Frankly I would use bipolar transistors for linear audio as they are far better in linearity if properly chosen. Earlier MOSFETs with poorer transconductance did make better amplifiers, but still not as good as well-chosen bipolar devices IMO.

David

[Stuart Polansky]

Thanks for the explanation and link, David. Very well said and explained.

Stuart

[DaveR]

YES, YES, YES - we have no banannas today. Sorry, had to do it.

I have read David's post and the article he had the link to. First, let me proclaim that I am not an expert at this stuff, actually not of anything I can think of. However, I do have a diminished capacity to read, listen and learn. So with that said...

If you consider a single device with, apparently, multiple MOSFETs on a single die bonded to a copper base - it seems that the individual transistors within the device would reach a current sharing equilibrium quickly. If they did not I'm not sure power MOSFETs would be a viable thing at all. Temperature differences between the individual transistors has to be at least pretty close given their proximity and the copper they are bonded to. Certainly no where near the 155-165 degree differences shown on the transconductance curves in the App Note.

When you consider using multiple devices in parallel it's pretty much the same thing. Though here you need to match the devices and use ballasting resistors to help them share the load, which we do. It seems to me the biggest difference between MOSFETs on a die and MOSFETs on a heat sink is the time it takes the temperature of one device to affect the others. We have found them to be stable and reliable as long as you don't short the wrong thing out with a scope or meter probe. Of course that won't happen unless you have the cover off and aren't being as careful as you should be.

Current production of power MOSFETs, at least that I'm aware of, are in fact designed for switching applications. The ones we used in the Mirror Image amps have low transconductance, about 2 S. They are Toshiba 2SK405 and 2SJ115. I've atached the transconductance curve graph for the 2SK405. We also used Hitachi 2SK414/2SJ119 initially, but their cost was considerably higher than the Toshibas. Neither has been made in years. You can find fakes on eBay, but they won't work well in an audio design. I purchased some a while back and took them to David's to see how they behave on his curve tracer. They behaved like something designed for switching. Recently I was lucky enough to find a stash of genuine Toshibas and purchased all of them.

I can't speak to David's feeling that bipolar would be a better choice than MOSFETs for an audio amplfier output. I do know that David and my old partner Ken are the two smartest guys I know when it comes to this sort of thing. Ken chose MOSFETs, I think they sound pretty darn good.

Attachments

2SK405Trransconductance.png (24.67 KiB) Viewed 21671 times

[DaveR]

I also meant to say that I would love to hear a Berning solid state amp.