Power supply simulator. / General Science and Electronics / Forums

Forums

4hv.org :: Forums :: General Science and Electronics

« Previous topic | Next topic »

Power supply simulator.

first previous 2 3 4 5 next last

Move Thread

LAN_403

Sulaiman

Mon May 07 2018, 11:39AM

Registered Member #162 Joined: Mon Feb 13 2006, 10:25AM
Location: United Kingdom
Posts: 3140

DerAlbi wrote ...

Usually the 150Â°C - 175Â°C limit arises from the plastic packaging which puts stress on the device due to different thermal expansion compared to the copper backplate.

I believe that the temperature limit is due to migration of aluminium from the interconnect/metallization layer(s) into the semiconductor, changing the doping level.
Many r.f. fets use gold metalization which allows typically up to 200 C before significant migration/diffusion occurs.
For short periods of time silicon can operate at much greater than 200C,
but as migration/diffusion rate increases massively with temperature,
the semiconductor device characteristics will also change quickly - permanently.
===================================== =========
Some rf power devices use (toxic in dust form) beryllium oxide between semiconductor and case to match the thermal expansion rates.
again to allow higher operating temperatures.
I'm not familiar with SiC thermal expansion coefficients,
but a quick google indicates that it is worse than pure silicon (4 vs 2.6 ppm/C) .
#######################################
sorry for straying from the main title but I hope it helps.

DerAlbi

Mon May 07 2018, 01:08PM

Registered Member #2906 Joined: Sun Jun 06 2010, 02:20AM
Location: Dresden, Germany
Posts: 727

AfAIK the annealing after fabrication of a semiconductor happens at much higher temperature (i have a number of 800Â°C in my head, nut sure).
I dont think the only problem is is mismatch between silicon and the rest - the copper backplate vs plastic packaging is enough to have a bi-metal-like behavior so the die suffers from mechanical stress. Mechanical stress changes electron mobility and therefore device properties [thats why if you mount TO220 or TO243 (all the big packages) a mounting torque is specified in the datasheet]. Many temperature cycles can lead to fatigue and crystal damage.

I never heard of aluminum causing problems and is replaced by gold. If gold is the reason for the high temperature tolerance.. i cant say no, but RF-devices are generally smaller volume higher price - a non standard plastic packaging might also do the job. There might be a lot of correlations but only a few causations. But it might be true.. cant find the aluminum problem through quick googling however

Sulaiman

Mon May 07 2018, 02:08PM

Registered Member #162 Joined: Mon Feb 13 2006, 10:25AM
Location: United Kingdom
Posts: 3140

From my era (1971)

Loads of more recent stuff via google.

Because high power rf transistors (I use gold metalized LDMOS) can tolerate higher temperatures,
hence much greater temperature differentials,
the encapsulation is a ceramic lid with a void between it and the semiconductor,
because as you rightly wrote, solid plastic encapsulation would cause failure,
AFAIK mainly due to wire-to-die bonding failure.
High power 'block' semiconductors usually use silicone resin for the encapsulant for the same reasons.

+1 on the need to not deform packages too much by over-tightening.

Ash Small

Mon May 07 2018, 04:19PM

Registered Member #3414 Joined: Sun Nov 14 2010, 05:05PM
Location: UK
Posts: 4245

DerAlbi wrote ...

There are no SOA curves for them

Since you gave some details like "UnitedSiC", 21A, 1.2kV... and i really couldnt believe such statement [SOA is a way too fundamental spec] Mouser has a datasheet.

On page 6 bottom right, you will find the SOA.
Have fun at 100V @ 1A @ DC @ 25Â°C and turn the current down to 500mA when the case [= backplate metal] temperature reaches 100Â°C because then only 60W dissipation is allowed as seen in page 6 bottom left.

Once they are too hot to touch, they are too hot

No. Human temperature sensing has nothing to do with semiconductor limits. The threshold you describe is about 50Â°C - unless you can show in the datasheet which specific device property derates at this point so that your design wont work, the temperature is fine. You can run with no issues at 120Â°C an more as long as you respect the specification. Usually the 150Â°C - 175Â°C limit arises from the plastic packaging which puts stress on the device due to different thermal expansion compared to the copper backplate. If you are careful within specs you can accidentally run a device until it desolders itself [during prototyping, and technically thats not in spec..]

Have fun building!

If you extrapolate down on the SOA graph, you'll see that as you approach 1 amp and 10 volts, you are into several seconds 'on' time. While it says the graph represents peak pulses, it doesn't specify how often those pulses can occur, therefore it is meaningless.

My interpretation of the preceeding graph is that at a case temperature of 50C it can dissipate over 100watts per device, that's 500 watts over five devices. My power supply only supplies 150 watts. What is wrong with my maths?

DerAlbi

Mon May 07 2018, 06:15PM

Registered Member #2906 Joined: Sun Jun 06 2010, 02:20AM
Location: Dresden, Germany
Posts: 727

You dont extrapolate, because the DC-Curve is already infinite long operation

1 or 10 seconds are between the 10ms and the DC-Graph! nothing to worry about.
Also repetition is not important here - see, the graph shows you how fast you have to switch at a given load. If you have an inductive load and simultaneously 400V @ 10A on the device you should switch faster than ~40Âµs. The device will survive and the mismatch in the internal structure of the device wiill not cause hot spots (with negative temp coeff) to fatally run away temperature wise due to heat capacity/conductivity of the individual hot spots.
Repetition wise, i see your problem, but thats what simulation is for: without knowing how bad the load behaves during switching one cant make an assessment there, so its not in the datasheet - resistive loads are mich nice to turn of than inductive loads, in reality you might have a mix of both and so on. Its up to the user to simulate/estimate switching losses. As long the individual switching action is within SOA and overall heat output (switching loss + conduction loss) is manageable you are fine.

Btw: Class-A means you have to follow DC-spec

Ash Small

Mon May 07 2018, 07:10PM

Registered Member #3414 Joined: Sun Nov 14 2010, 05:05PM
Location: UK
Posts: 4245

DerAlbi wrote ...

You dont extrapolate, because the DC-Curve is already infinite long operation

I'm working on the premise that with a power supply giving 25-30 V, I'll have about ten volts across the switch (50% of what's left after biasing the gate). This is the average voltage. The device has an open resistance 0f 80 mOhm, so with ten volts drain-source, and, for argument's sake, 1 amp current, internal resistance of the device needs to be biased at 10 Ohms, with a ten Ohm drain resistor and ~5-10 Ohm source resistor. Given this scenario, I'm guessing if the load is connected between drain and source, it should also be ~ 10 Ohm load. With five devices in parallel, load would want to be two Ohms.

Ten volts into two Ohms = 5 Amps, 10 V x 5 Amps =50 Watts.

I'll be happy if I can achieve half this (25W).

It's just a question of experimenting with bias until they are warm

I've also done some calculations regarding clamping the current with yet more chokes (parafeed style). A typical tube amp will have a smoothing cap of 80uF and a 40 H choke, which gives a resonant frequency just under 3 Hz.

With a 75200uF smoothing cap (final stage) and a 40 mH choke the resonant frequency is just under 3Hz, identical to the tube circuit. With only 10mH choke on 75200uF, res. frequency is only 6Hz. I have sufficient cores for at least 10mH, which should be sufficient at least for now.

Have I made any obvious mistakes in the above assumptions?

Sulaiman

Mon May 07 2018, 10:19PM

Registered Member #162 Joined: Mon Feb 13 2006, 10:25AM
Location: United Kingdom
Posts: 3140

I think that you have made a significant mistake :
you are using a high voltage transistor where it would be better to use one with a more suitable rating.
This is because, in general, an increased voltage rating means:
- higher cost
- higher ON resistance
- lower frequency response

Of course, for learning purposes none of the above are important.

I've not had much experience with thermionic valves but these jFET SiC transistors look to have very similar characteristics.
As I think that you should beusing a higher supply voltage for these transistors,
you could look at old designs using thermionic triodes.

Ash Small

Mon May 07 2018, 11:51PM

Registered Member #3414 Joined: Sun Nov 14 2010, 05:05PM
Location: UK
Posts: 4245

Sulaiman wrote ...

I think that you have made a significant mistake :
you are using a high voltage transistor where it would be better to use one with a more suitable rating.
This is because, in general, an increased voltage rating means:
- higher cost
- higher ON resistance
- lower frequency response

Of course, for learning purposes none of the above are important.

I've not had much experience with thermionic valves but these jFET SiC transistors look to have very similar characteristics.
As I think that you should beusing a higher supply voltage for these transistors,
you could look at old designs using thermionic triodes.

I look at it slightly differently, yes, it's a triode, solid state, but works exactly the same as a tube/valve triode in Class 1 (voltage controlled).

It's designed for high frequency switching, so plenty fast enough for audio. It's 'on' resistance is 80 milliohms. They weren't cheap, but they're limited production and high power rating.

I've used 'modern' fast switching JFET's for audio before, 500mA ones which run really happily around 20-50mA, and I've run them at up to 350mA for short periods of time.

These 'modern' JFET's are much more predictable than older types, the production techniques have advanced considerably, yet people still tend to use the old J201 and a few others from around the same period. I don't expect single ended Class A amps to ever make a comeback though, because of the perceived poor efficiency.

These SiC power JFET's should be good for the same voltages that thermionics usually run at, but these are capable of much higher currents than thermionics, so can be run at much lower impedences, which eliminates the need for all that copper and iron in the output transformer.

It does mean a considerably larger output capacitor, but these are much easier to come by these days than back in the heyday of singled ended Class A amps, the same goes for the electrolytics in the power supply.

The mass of the iron and copper in the chokes is a 'per watt' thing, so an amp of this general design of a given power output needs the same mass of copper and iron whether it's a thermionic circuit running 300V and 100mA or low impedence solid state running 30V and 1A, or whatever the figures are. At least, that's my impression.

I do have a huge Parmeko power transformer, several 40H and 10H chokes, and a huge Williamson output transformer, but I also have a load of 807's I was planning to use with them.

The basic idea was to run the SiC JFET's at a low enough impedence not to need an O/T, although this does require a big output capacitor

I've finished the first power supply choke, by the way. 116 turns of PVC insulated 7 core 2.5mm^2 on a T300-26D micrometals core, giving 2.3mH. I'd have been happy with 2mH and was aiming for 2.2mH, so I'm pretty pleased.

I do need to get some more 7 core 2.5mm^2 wire, though.

Sulaiman

Tue May 08 2018, 03:48PM

Registered Member #162 Joined: Mon Feb 13 2006, 10:25AM
Location: United Kingdom
Posts: 3140

Not that it matters in this case, just FYI,
the general consensus is that the most efficient magnetics are when core loss (mainly due to B-H hysteresis) is approximately the same as wire losses (mostly I^2.R)
PVC is a good insulator;
- excellent when high voltages are present
- terrible for getting heat out of the wire.

I'd be more inclined to look for some suitable enameled copper wire.

Ash Small

Tue May 08 2018, 06:26PM

Registered Member #3414 Joined: Sun Nov 14 2010, 05:05PM
Location: UK
Posts: 4245

Sulaiman wrote ...

Not that it matters in this case, just FYI,
the general consensus is that the most efficient magnetics are when core loss (mainly due to B-H hysteresis) is approximately the same as wire losses (mostly I^2.R)
PVC is a good insulator;
- excellent when high voltages are present
- terrible for getting heat out of the wire.

I'd be more inclined to look for some suitable enameled copper wire.

I have some 14 SWG enamelled wire, but felt I had a sufficiently large window to use the PVC coated stuff.

That's the next thing to test, I guess, see if it gets warm with 5 Amps through it,

The PVC gives more protection than enamel, I think, plus I didn't want to completely fill the window with copper. It's only about 50% copper, actually significantly less.

You can never have too much iron and copper

EDIT:- Unfortunately I still can't measure the DC resistance of choke, as the only meter I have available at present won't measure less than 0.5 Ohm.

first previous 2 3 4 5 next last

Moderator(s): Chris Russell, Noelle, Alex, Tesladownunder, Dave Marshall, Dave Billington, Bjørn, Steve Conner, Wolfram, Kizmo, Mads Barnkob