Lateral MOSFET torture...

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I did a fun little experiment with a Hitachi 2SJ83 lateral mosfet a couple of weeks ago. It is probably the same die as used in the 2SJ50, the parameters in the datasheet look the same.

Mounted on the small "heatsink" (~4degK/W) shown in the picture I gave it 50V drain to source and then applied 10V to the gate with a 330 ohm gate resistor to suppress oscillation. The current after a short short while was 6A. (300W dissipation in the 100W rated part) The fan in the picture was NOT running.

Can anyone guess:
a) How long it survived?
b) How hot the "heatsink" got?

--
Joakim :D
 

Attachments

  • tortyr.jpg
    tortyr.jpg
    96.6 KB · Views: 1,812
I didn't have any fast responding instruments for measuring the current, I just used the ones on the power supply. So it's 6A after maybe 0.2 seconds or something. The initial current might have been greater, but the laterals conduct less current at the same Vgs the hotter they get...

The resistance of that "thin wire" is 6 milliohms so it is probably not a problem.
 
My guess is maybe 5 minutes, with a heat sink temp of 150C.
Mosfets will take junction temps of around 300C before they die instantly. I've tortured a few hexfets in my day - with the epoxy package split open, glowing orange inside, but still beahving as an amplifier. Didn't work after it was turned off, cooled down, and turned back on, however.
 
Glowing orange and still behaving as an amplifier? :hot: I'm not surprised it didn't work afterwards though! The lateral mosfet in my experiment does something interesting though:



The device under test is a lateral mosfet, a Hitachi 2SJ83, a 2SJ50 equivalent in a weird plastic package. It is mounted on the small "heatsink" (~4degK/W) shown in the picture in the first post. I gave it 50V drain to source and then applied 10V to the gate with a 330 ohm gate resistor to suppress oscillation. The current after about 0.1 seconds was 6A. (300W dissipation in the 100W rated part) The fan in the picture was NOT running.

The current starts decreasing very rapidly, maybe to 2 amps after a second. (the current has a negative temperature coefficient above 100mA) The heatsink continues to heat up. This negative temperature coefficient is not enough to limit the temperature much though due to de woefully inadequate heatsink. :D

Next, the FET starts to draw current into the gate, thereby lowering the current through it and the dissipation. When the current going into the gate exceeded 8mA a small pop was heard, the gate voltage dropped to 1.2V (drawing 25mA then), the drain current dropped to 0.2A and the heatsink started cooling off. After turning off the drain-source power supply the gate was still drawing current.

By turning off the gate supply the transistor stops drawing gate current of course, but when turning on the gate supply it doesn't draw any current. :confused: Maybe it wasn't destroyed? I did the same experiment a couple of times, and it does the same thing each time :eek:

If I drop the gate voltage so the gate current never exceeds 8mA I can get the heatsink to stabilize at 165 degrees celsius with 40W of power dissipation. The transistor is mounted without insulating washer and the temperature is measured right below it so the die temperature seems to never want to exceed 200 degrees by much.

It is interesting that it has this built-in overtemperature protection, no wonder they got the reputation of being rugged and outlasting fuses! But I wonder, what is it that causes this? Is it something with the built-in gate protection zeners maybe? The thyristor effect when exceeding 8mA of gate current is very interesting :confused:
 
I had always doubted that external zeners plus fuses would be enough to protect these FETs but I happened to run across these usenet posts by Phil Allison (same guy that wrote the VI-limiting article on the ESP pages):
http://groups.google.com/group/rec.audio.tubes/browse_frm/thread/65ec8f2081b5e1e2/d7b54f59a162286e
http://groups.google.com/group/rec.audio.tubes/browse_frm/thread/65ec8f2081b5e1e2/33f891609d862346
http://groups.google.com/group/rec.audio.tubes/browse_frm/thread/65ec8f2081b5e1e2/dd74bc5db48967ed

I had a hard time believing that there really is a semiconductor component that will outlast fuses with such simple protection so I had to try it. :) Especially the part about mounting bolts working loose without meltdown...

http://groups.google.com/group/rec.audio.tubes/msg/84eaa3a11d8efada

He thinks the zeners must be chosen wisely. But, just think of what happens if the mosfet is cool and then gets hit by 14V limited by the gate protection zeners. I'm pretty sure the thermal time constant of the leadout wire is shorter than the one of the die, so I guess the drain leadout wire fuses in short order at the current which ensues...
 
Lateral FET Torture

It should not be surprising that one can destroy a lateral FET with an overcurrent condition. Even if the FETs have external protection zeners, they will not clamp at a sufficiently low voltage to prevent overcurrents from potentially destroying the FETs. Typically the zeners are selected to prevent gate breakdown, and this voltage limit is around 12-14V, more than enough to drive the FETs past their SOA.

The real question is how fast any overcurrent protection circuit needs to respond to prevent damage to the FET. Over short intervals an audio amp power supply can provide tens to hundreds of joules of energy, so we cannot count on the power supply providing limiting. One way of looking at the situation is to consider the thermal time constant of the device vs. that of the protection circuit. In the case of a fuse the thermal time constant for a bondwire will be less than that of a fuse, since the bondwire is physically much smaller. It is possible to mitigate this problem by paralleling many output devices, but this adds capacitive loading and cost.

IMO, the best approach is to employ an active overcurrent circuit (monitoring the current delivered from both the positive and negative rails) and to set the current trip point somewhat less than the current SOA of however many devices are paralleled. The amp I am designing uses slaved overcurrent monitor circuits that remove gate drive to the output devices if an overcurrent condition (>80 amps) is detected on either the pos or neg rails. Once the overcurrent circuit is tripped, the output device gate drive remains off until the amplifier is power cycled. The time from overcurrent detect to gate turn off is about 10 usec. Since the overcurrent condition cannot occur for loads less than 1 ohm, the amp should never trip unless its output is shorted or is driving an unrealistically low impedance.
 
The Hafler DH500 had three pair of 8A Hitachi in parallel, 10A fuse.

The DC off-set was enough to blow this fuse into a short on the output jacks, ±93V rails.

The only time you could blow the outputs was with sustained clipping into 2R. The external 10V zeners would blow, and then the gates would follow.

I had one with a blown bias transistor, biased itself into class A. I had up-graded the stock 35CFM fan to 110CFM, so it would run for about five minutes before it would overheat. Did this many times before I figured out what the problem was. That was 1982, the amp is still in service in 2007 in a friend's PA.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.