Cheezeball DC Load: DL24P: Pump, or Dump ???

[deuteron]

In principle, mounting the Mosfet separately on the cooler is a good idea and may be - depending on the Mosfet package - even necessary. The only problem with this is that the thermal sensor controlling the fan speed must be transferred to the cooler as well. But this is certainly possible if the kind and parameters of the internal sensor are known. Of coarse one can make it also simple and just connect the fan to the power supply line so that it runs full speed whenever the load is in operation. My testing is done solely with a power supply with current limiting so far just to avoid any damage of the board. Nevertheless I'm planning to install a fuse either, on the board or just simply within the current leads to the load.
Today the linear Fets have arrived. I'll install one tomorrow and continue my testing.
Best regards
Stefan

[Pukker]

The only problem with this is that the thermal sensor controlling the fan speed must be transferred to the cooler as well. But this is certainly possible if the kind and parameters of the internal sensor are known.

The external sensor is NTC is 10k Beta 3435,
the internal sensor is also 10k, don't know if the Beta is the same.

[vehiculeselectriques]

Hello,

Mine is mounted with 1 FDL100N50F (SOA DC60V40A-200V10A 2500W 150°C)
with 250W tdp 6 heat tube good CPU cooler i have 300W48V without trouble,
diode go very hot though !!!
it would go higher power by tripling diode number but ok as is for my usage.

[deuteron]

The only problem with this is that the thermal sensor controlling the fan speed must be transferred to the cooler as well. But this is certainly possible if the kind and parameters of the internal sensor are known.

The external sensor is NTC is 10k Beta 3435,
the internal sensor is also 10k, don't know if the Beta is the same.

That's good to know. Do you also know, which are the temperatures the beta value is related to (e.g. 25°C/85°C) ?

Hello,

Mine is mounted with 1 FDL100N50F (SOA DC60V40A-200V10A 2500W 150°C)
with 250W tdp 6 heat tube good CPU cooler i have 300W48V without trouble,
diode go very hot though !!!
it would go higher power by tripling diode number but ok as is for my usage.

I just had a look into the data sheet of that Mosfet. It is extremely powerfull, amazing that this is possible within the standard TO-264 package ! I would directly switch to this model, if it was explicitly a 'linear' device. But it isn't. Perhaps it's running fine though, because in normal operation one is very far away from the SOA limit lines.

[jtr1962]

2)  The FET will promptly explode.  The problem outlined by Syonyk (https://syonyk.blogspot.com/2018/06/the-atorch-purple-fan-mosfet-destroyer.html) still exists.  I think the problem is an opamp stage driving the fet:  it is bode unstable, and the oscillations exceed the ±20v Vgs spec.     My fet failed short circuit - which will cause big problems if connected straight to a Lipo battery (it's funny that the brand name is "ATORCH" - this should be taken literally).

I just ordered one of these (DL24 actually, not DL24P) but it won't arrive until mid July. Anyway, having a look at the schematic I think the problem is the low-pass filter to the integrator op-amp consisting of R6 and C5. I've built many constant current sources using op-amps. I never put a low-pass filter on the current feedback to the op-amp. That creates another pole, which likely is the cause of the instabilities mentioned. Quick fix is to simply remove C5. If anyone here who has a DL24P can try that, please let me know if it works.

In layman's terms we want any fluctuations in the current to feed back into the integrator op-amp right away, instead of being delayed by the low-pass filter. If they're delayed, the integrator response will lag, possibly leading to the harmful oscillations.

Back to back 10V to 15V zeners between the gate and source is a good failsafe regardless, even if removing C5 fixes the instability issue.

[Pukker]

That's good to know. Do you also know, which are the temperatures the beta value is related to (e.g. 25°C/85°C) ?

Sorry, don't know.

[deuteron]

2)  The FET will promptly explode.  The problem outlined by Syonyk (https://syonyk.blogspot.com/2018/06/the-atorch-purple-fan-mosfet-destroyer.html) still exists.  I think the problem is an opamp stage driving the fet:  it is bode unstable, and the oscillations exceed the ±20v Vgs spec.     My fet failed short circuit - which will cause big problems if connected straight to a Lipo battery (it's funny that the brand name is "ATORCH" - this should be taken literally).

I just ordered one of these (DL24 actually, not DL24P) but it won't arrive until mid July. Anyway, having a look at the schematic I think the problem is the low-pass filter to the integrator op-amp consisting of R6 and C5. I've built many constant current sources using op-amps. I never put a low-pass filter on the current feedback to the op-amp. That creates another pole, which likely is the cause of the instabilities mentioned. Quick fix is to simply remove C5. If anyone here who has a DL24P can try that, please let me know if it works.

In layman's terms we want any fluctuations in the current to feed back into the integrator op-amp right away, instead of being delayed by the low-pass filter. If they're delayed, the integrator response will lag, possibly leading to the harmful oscillations.

Back to back 10V to 15V zeners between the gate and source is a good failsafe regardless, even if removing C5 fixes the instability issue.

That might be a very important observation. Personnaly I didn't design current sources so far, but I did a handfull of other regulators for physics experiments. This kind of feetback coupling is also new to me. First I would have to study the oscillation by my one. I didn't connect a scope to the gate so far. Ill see whether I have the time to do so in the next days. Then we will see what happens, if C5 is removed.

[deuteron]

Another interesting observation from the DL24(P) circuit diagram: Why do they waive the pull down resistor at the mosfet gate ? As far as I remember with earlier models there have been some accidents with connecting the power supply / battery to the load before it was powered on. Perhaps a permanently open gate ?

[jtr1962]

Another interesting observation from the DL24(P) circuit diagram: Why do they waive the pull down resistor at the mosfet gate ? As far as I remember with earlier models there have been some accidents with connecting the power supply / battery to the load before it was powered on. Perhaps a permanently open gate ?

Normally that wouldn't be an issue if the loop was stable. The MOSFET would be turned on in the absence of a battery but the opamp would respond fast enough once the battery was connected to bring it back to the linear region before any damage was done. Or at least that's my experience with the circuits I've made. I often set the current prior to connecting the battery. No issues with MOSFETs blowing once I connected the battery.

That said, it's probably better to set the current to zero before connecting a battery. Or perhaps the device itself could detect if a battery is not connected (i.e. zero volts on the A+ and V+ terminals), and keep the current control at zero until such time as it detects a battery voltage. Once it does, it can ramp up to the set current. If it doesn't already do this, the next version of the firmware should incorporate this fix. The idea is to make it as failsafe as possible.

Side note: If anyone wants to try a power upgrade, I found a good candidate for a replacement MOSFET:

https://www.mouser.com/ProductDetail/747-IXFH94N30P3

300V drain-source breakdown voltage, 94 amps maximum current, 1.04 kW maximum power dissipation

And it's relatively inexpensive at only $12.99. I imagine with water cooling you might be able to reach 1 kW.

[deuteron]

Normally that wouldn't be an issue if the loop was stable. The MOSFET would be turned on in the absence of a battery but the opamp would respond fast enough once the battery was connected to bring it back to the linear region before any damage was done. Or at least that's my experience with the circuits I've made. I often set the current prior to connecting the battery. No issues with MOSFETs blowing once I connected the battery.

Ok, just to be sure that I got you correctly. Say without a battery, the Mosfet gate can be in some undefined state, i.e. perhaps it is somewhat opened up. Then, when connectiing the battery, the regulator would close the gate since there is no positive set-voltage at the op amp. In this way a strong current through the Mosfet would be avoided, if this 'coming to life moment' of the op amp is fast enough, right ? On the other hand, I think that adding a pull down resistor to the gate (~ 100kOhm) would improve the safety anyway. Btw it's pretty similar to the grid resistance in the case of tube valves. Here the extra resistance is important to avoid some static charge of the grid. Furhermore it helps preventing oscillations of the tube.

Side note: If anyone wants to try a power upgrade, I found a good candidate for a replacement MOSFET:

https://www.mouser.com/ProductDetail/747-IXFH94N30P3

300V drain-source breakdown voltage, 94 amps maximum current, 1.04 kW maximum power dissipation

And it's relatively inexpensive at only $12.99. I imagine with water cooling you might be able to reach 1 kW.

I just had a look to the data sheet. Firstly I'm really surprised about the package size, it's only a TO-247, just like the original one. This is a strong hint that the actual (long term) heat dissipation ability is strongly limited. Secondly it is not meant for linear applications. And lastly, if you look at the SOA diagram, then you find the last limiting line is the one for 1msec pulse duration. Since there is even no 10msec line (not to mention DC), from my limited experience, I'm pretty sure, that DC operation wouldn't work.
I'm just testing the linear IXTK60N50L2 and it seems to be good for about 400W in DC (up to 400V). But this won't be the power I'll go up to, because the cooling power of my CPU cooler seems to be limited to about 150W. Neverthelesss, that's fine for my application. What I want is a stable device, which max power is given ONLY by the coolers maximum heat dissipation ability.

[jtr1962]

Ok, just to be sure that I got you correctly. Say without a battery, the Mosfet gate can be in some undefined state, i.e. perhaps it is somewhat opened up. Then, when connectiing the battery, the regulator would close the gate since there is no positive set-voltage at the op amp. In this way a strong current through the Mosfet would be avoided, if this 'coming to life moment' of the op amp is fast enough, right ? On the other hand, I think that adding a pull down resistor to the gate (~ 100kOhm) would improve the safety anyway. Btw it's pretty similar to the grid resistance in the case of tube valves. Here the extra resistance is important to avoid some static charge of the grid. Furhermore it helps preventing oscillations of the tube.

That's the general idea, avoid destructive currents through the MOSFET when you connect a battery.

Thinking about this some more, another way to accomplish the same thing is to simply clamp the maximum g-s voltage via a zener diode. If you look at MOSFET data sheets, generally low g-s voltages result in current limiting. For example, since this is a 20 amp tester, you could clamp the g-s voltage at a value that limits the drain current to ~30 amps. Even if a battery is connected, no more than 30 amps would flow, but the driving op-amp would quickly bring that down to the set current. With the stock MOSFET, limiting Vgs to about 5.5V would keep the maximum drain current in the 30 to 40 amp area.

I just had a look to the data sheet. Firstly I'm really surprised about the package size, it's only a TO-247, just like the original one. This is a strong hint that the actual (long term) heat dissipation ability is strongly limited. Secondly it is not meant for linear applications. And lastly, if you look at the SOA diagram, then you find the last limiting line is the one for 1msec pulse duration. Since there is even no 10msec line (not to mention DC), from my limited experience, I'm pretty sure, that DC operation wouldn't work.
I'm just testing the linear IXTK60N50L2 and it seems to be good for about 400W in DC (up to 400V). But this won't be the power I'll go up to, because the cooling power of my CPU cooler seems to be limited to about 150W. Neverthelesss, that's fine for my application. What I want is a stable device, which max power is given ONLY by the coolers maximum heat dissipation ability.

It's possible DC might not work but the IXTK60N50L2 sounds like a good candidate. I personally have no need to go above even 150 watts as I would be mostly testing single cells.

[deuteron]

Thinking about this some more, another way to accomplish the same thing is to simply clamp the maximum g-s voltage via a zener diode. If you look at MOSFET data sheets, generally low g-s voltages result in current limiting. For example, since this is a 20 amp tester, you could clamp the g-s voltage at a value that limits the drain current to ~30 amps. Even if a battery is connected, no more than 30 amps would flow, but the driving op-amp would quickly bring that down to the set current. With the stock MOSFET, limiting Vgs to about 5.5V would keep the maximum drain current in the 30 to 40 amp area.

It's possible DC might not work but the IXTK60N50L2 sounds like a good candidate. I personally have no need to go above even 150 watts as I would be mostly testing single cells.

It seems to be that you and me we are unrevaling a dialog here. It has been said almost everything about these loads in the last years. But only almost ;-) I have the feeling that your application is rather low voltage but high current, right ? I think for that it is absolutely necessary to strengthen both the 'polarity diode' and the current guiding tracks on the PCB. From my guessing, I would say the original state is ok for currents up to 5 amps, but not much more. For me it's fine to stay below 150W either, but also my current needs are rather low. As written several times, I want to have a reliable device but without any magic smoke. I think the idea to limit the gate voltage with a suitable Zener is a really good one. Actually I'm using two 15V Zeners back to back to save the Mosfet from any instabilities. As soon as I have the time, I will connect a scope to the gate and look for the regulation behaviour. I'm not sure for the moment, did you made the comment about the extra cap in the feedback channel ? Anyway, this is indeed something, which should be clarified. Perhaps the instabilities can be cured in this way. The operation voltages of the regulator is in the range of 3.5 to roughly 5 Volt with the above Mosfet for the currents and voltages I used so far. So instead of the 15V one can probably switch to a much lower Zener voltage and thus solving the overcurrent issue at the same time.
Actually I make a whole series of temperature measurements under various heat loads. So far I tested the IXTK60N50L2 up to 160W at 80V. For the low voltage / high current regime, I need a more powerful supply (currentwise) which I do not have at home. I hope I can do these measurements in our physics lab tomorrow. As soon as the measurements are completed I will post the results here.

[jtr1962]

It seems to be that you and me we are unrevaling a dialog here. It has been said almost everything about these loads in the last years. But only almost ;-) I have the feeling that your application is rather low voltage but high current, right ? I think for that it is absolutely necessary to strengthen both the 'polarity diode' and the current guiding tracks on the PCB. From my guessing, I would say the original state is ok for currents up to 5 amps, but not much more. For me it's fine to stay below 150W either, but also my current needs are rather low.

Right now most of my testing will be low voltage and 5 amps are less. However, I want the option to test at higher currents should I need it in the future. Correct that I'll seriously consider upgrading the tracks and polarity diode for that. In fact, I might consider ditching the polarity diode altogether as I'm very careful when I'm hooking up a battery. Or consider using a P-channel MOSFET for reverse polarity protection:

https://hackaday.com/2011/12/06/reverse-voltage-protection-with-a-p-fet/

I might even see if I can go well past 20 amps with some modifications.

In the future I might be interested in testing things like 4 series or 8 series LiFePO4 batteries at fairly high currents. That's where a power upgrade would be needed. For now though 150W seems sufficient.

As written several times, I want to have a reliable device but without any magic smoke. I think the idea to limit the gate voltage with a suitable Zener is a really good one. Actually I'm using two 15V Zeners back to back to save the Mosfet from any instabilities. As soon as I have the time, I will connect a scope to the gate and look for the regulation behaviour. I'm not sure for the moment, did you made the comment about the extra cap in the feedback channel ? Anyway, this is indeed something, which should be clarified. Perhaps the instabilities can be cured in this way. The operation voltages of the regulator is in the range of 3.5 to roughly 5 Volt with the above Mosfet for the currents and voltages I used so far. So instead of the 15V one can probably switch to a much lower Zener voltage and thus solving the overcurrent issue at the same time.

Yes, I'm the one who suggesting removing C5. It should hopefully fix the instability issue. And I'll probably use back-to-back 5V Zener diodes to limit Vgs to a value which limits current to 30 or 40 amps.

Actually I make a whole series of temperature measurements under various heat loads. So far I tested the IXTK60N50L2 up to 160W at 80V. For the low voltage / high current regime, I need a more powerful supply (currentwise) which I do not have at home. I hope I can do these measurements in our physics lab tomorrow. As soon as the measurements are completed I will post the results here.

OK, thanks. I look forward to seeing your results. I plan to do testing and look for oscillations with my scope once my load arrives.

[deuteron]

Today I in fact had some time for measuring. I'll divide the upcoming postings in two subjekts.

Firstly: My findings concerning the so called 'ringing' at the gate:

I connected a digital oscilloscope to the gate-source contact of the Fet, which are bridged by two 15V back-to-back Zeners. With this I just followed some recommendations here from the board. I don't beleive that this was interfering with my measurements. Since I learned that the 'strange' behaviour of the regulator is strongest, when the Fet is wide open, I connected the load to our hi current magnet power supply. I used only some volts but up to 10 amps to arouse oscillations if there are any. What I saw I would call some kind of 'knocking' rather than 'ringing' ;-) Please look at the photographes to get an idea what I'm meaning with this. The first picture shows the 'worst case' I was able to initiate by switching the current from 0A to 10A via the start/stop button. The voltage was only 3V to open up the Fet as widely as possible. You see that the regulator overshoots the desired value by more than 4V and then goes back to the target value of around 6V. In the second picture you see the same behaviour but only at 5A, 2V. It's much less harmfull and should be ok during normal operation. For the moment I don't have any idea, where this is coming from. The mentioned C5 ? The third picture shows the structure of the constant gate voltage during operation at 5A. It looks the same as with the gate closed. The frequency is around 125MHz overlayed by some beating with a tenth of the basic frequency. Just to be sure that this behaviour has nothing to do with the Zener bridging, I'll remove these tomorrow and measure once again. The IXTK60N50L2 should be strong enough to survive those excess voltages. Vgs max is 30V or 40V for continuous or transient, respectively.
I think I found C5 on the PCB. Wow, is that tiny !!! Removing it may not be a big problem, but no chance for a repair. At least not by me.

Small addendum: The behaviour of the regulator when switching back to 0V was absolutely clean, just a single step and ready !

[deuteron]

Second part: DL24P with Noctua Nh-9Li https://noctua.at/de/nh-l9i

Modifications:
1) Replacement of the original cooler with the above mentioned one (TDP ~ 100W). Thermal compound was Artic MX-6.
2) Thermal connection of the reverse polarity diode to the heat sink body (see picture)
3) Two 15V Zener in back to back configuration for protection of the Mosfet gate in both directions (In the meanwhile I realized that only the plus polarity may be necessary.)
4) Exchange of the (fake ?) IRFP264 by the IXTK60N50L2 linear Mosfet (SOA diagram with power dissipation lines see picture)

Set up:
I didn't do any modification to the PCB itself so far. No strengthening of PCB traces, the reverse polarity diode is still the original one. I measured both case temperatures, the ones of the Mosfet as well as those of the (thermally connected) diode. For the Mosfet temperature I used the external temperature sensor of the load mounted within the mounting hole of the Fet with thermal compound. For the diode case temperature I used my Fluke DMM with thermal sensor connected to the case with thermal compound. See pictures and diagrams below.

Results:
1) The case temperature diagram shows a nice linear dependence of the temperature from the dissipated power with a thermal coefficient of 0.276°C/W. The arrangement is stable up to at least 160W, which means about 70°C case temperature. Perhaps one may extent it to 180W but somewhere here is the limit with the CPU cooler used.
2) The connection of the reverse polarity diode to the heat sink body seems to be of some advantage. The case of the diode stays below 70°C for power values up to 150 W. For voltages of 30V and below a current of 5A is possible, which is a conservative guess of what the PCB traces can withstand.

Summary:
The decribed arrengement allows a usage of up to 150W of power dissipation. Voltages of 30V and below allow a current up to 5A (PCB limitation, perhaps a bit more). For voltages from more than 30V the current limit is given by the max heat dissipation. For me and my applications that's fine and I will stick to it. Nevertheless I ordered another DL24P PCB board and I will do some further experimentation with it, particularly concerning the regulator and it's stability. There is another point, which is is not perfectly solved. Especially low currents are not very stable (+/- 2 up to 3mA). This I think is also a regulator issue. Hopefully there is some improvement possible.

[jtr1962]

Looking at the first part, the overshoot in the step response going from 0 to 10A could indeed be due to C5. Same thing with the oscillations. The back to back 15V Zener diodes are probably making no difference in the response because Vgs isn't high enough to turn them on. I plan to try similar experiments when mine arrives. First see under what conditions I get oscillations, then see if removing C5 fixes that.


On the second part:

1) Probably a good idea to keep both polarities of Zener diodes. It can't hurt, but it will clamp any negative oscillations.
2) Good idea connecting the reverse polarity diode to the heat sink.
3) Thermal compound is very important here given the small surface area of the MOSFET-heat sink interface. From my experience working with Peltiers, a poor thermal interface can result in a 5°C or more temperature differential between the device attached to the heat sink and the heat sink.
4) For greater reliability I agree about replacing the stock MOSFET.
5) If I need to go much over 150 watts I would consider mounting the MOSFET on a heat sink remotely. I have some heat sinks I've used for Peltier projects which have a thermal cofficient of around 0.05°C/W when used with a 160mm fan. Even for 1000 watts load, that's only a temperature rise of about 50°C above ambient.
6) 70°C is well below the absolute maximum operating temperature of the MOSFET and reverse polarity diode. That implies reliable operation.
7) The instability at low currents may be caused by C5, or perhaps by the integrator having too low a time constant. I'll examine this part of the circuit when my device comes.

Like you, for now I'm more interested in a reliable, stable, accurate load than I am about increasing TDP.

[beanflying]

Darn EEVBlog threads that keep popping up reminding me of jobs I need to tidy up. 

Several years ago my OG 150W load 'loaned' the fairly junky stock twin fanned cooler to the 'fancy' DL24P and got mothballed to 'the shelf'. Due to a recent shuffle of some PC hardware and adding a big boy cooler to the 5900X in the stack I finished up with a spare AMD Prism.

I will load up the DL24P when it gets the Cooler on it and some time and see how it goes for temps.

Then all it needs is someone with a Laser Cutter and several 3D printers to design and make an enclosure for them in his 'spare time' 

[deuteron]

Third part: Further systematic testing
Today I already got the new bare PCB ordered from Aliexpress only some days ago ! At least it was less than 1 week from China to Germany, surprising ! Most of the following exercises I did with the new (original) board, i.e. with another copy of the chinese IRFP264. Since I didn't use any cooler, the on time was restricted to some seconds at most. But for this kind of experimenation (step response) this is just fine.

1. Overshooting:
Below 5A and this even at a very low Vds 0f 2V there is no overshooting at all. We see some action of the regulator, but it's working just fine. See picture No.1 for example. From 10A on (3V, 10A to open up the Mosfet widely) the overshooting was getting more and more pronounced. Pictrure No.2 shows the step response from (3V, 0A) to (3v, 15A). Obviously the regulator stops overshooting at 11V, which seems to the maximum possible regulator voltage. I never saw any gate voltage larger than 11V. From that in principle there is no Zener blocking needed. Nevertheless, for the use at 5A max a 6V Zener may be fine just to prevent extremely high currents during the onset of the regulation.

Oszillations:
Next was to look once again at the high frequency osziallations (picture No.3). These very high frequencies (>100MHz) are probably environmental noise. It strongly depends on whether I connected the measuring head of the scope directly at the Fet legs or with some centimeters of unshielded cabling in between. So, I don't think that it originates from the regulator cicuit. The amplitude is only +/- 10 mV and the Fet can't follow these frequencies anyway. I then went up with the time scaling to 200 ms, equivalent to some Hz. Here we see the origin of the unstable current reading: picture 4. The behaviour is most prominent at very low current settings, i.e. some 10 mA. Interestingly there are some settings (more or less equidistant every ~40mA), in which the reading is almost completely stable corresponding to a mostly flat line on the scope (picture 5). There are even completely flat examples !

Conclusions:
My impression without being an expert in electronics is that this kind of jumps in Vgs is a digital artefact. Perhaps the reason can be found in the way the regulator voltage is generated as a DC transformation of a PWM signal. Once again perhaps the PWM filter they put before the buffer isn't efficient enough in producing a flat DC current (prolongation of RC time constants ?). Unfortunatly I can't append the schematics here, I'll try it in a further post below. Concerning the 'C5 problem' I'll ask our electronics engineer to unsolder it since it is far to small to be done by hobbyist means. I agree that the overshooting problem may be caused hereby. Let's see, I will report. So far so good, I think we are on a good way to make this thing a reliable device.

[deuteron]

Here come the well known schematics once again.

[deuteron]

1) Probably a good idea to keep both polarities of Zener diodes. It can't hurt, but it will clamp any negative oscillations.
2) Good idea connecting the reverse polarity diode to the heat sink.
3) Thermal compound is very important here given the small surface area of the MOSFET-heat sink interface. From my experience working with Peltiers, a poor thermal interface can result in a 5°C or more temperature differential between the device attached to the heat sink and the heat sink.
4) For greater reliability I agree about replacing the stock MOSFET.
5) If I need to go much over 150 watts I would consider mounting the MOSFET on a heat sink remotely. I have some heat sinks I've used for Peltier projects which have a thermal cofficient of around 0.05°C/W when used with a 160mm fan. Even for 1000 watts load, that's only a temperature rise of about 50°C above ambient.
6) 70°C is well below the absolute maximum operating temperature of the MOSFET and reverse polarity diode. That implies reliable operation.
7) The instability at low currents may be caused by C5, or perhaps by the integrator having too low a time constant. I'll examine this part of the circuit when my device comes.

Like you, for now I'm more interested in a reliable, stable, accurate load than I am about increasing TDP.

to 1) See my new results above
to 2) Yes, it helps particularly when the heat sink is relatively cold (low overall wattage)
to 3) Yes, I tried to use one of the best ones available. It seems to me I killed one real (not fake) Mosfet because of bad thermal connection to the heat sink, i.e. there was not enough thermal compound.
to 4) Yes
to 5) The problem here is the thermal sensor which steers the fan. It can't be easily tranferred to the remote heat sink. On the other hand, connecting the fan to the supply voltage and let it always run may be fine too.
to 6) Yes, 70°C is a conservative but reasonable limit. It is still not too hot to get your fingers burned when touching it. Additionally the internal temperature of the Fet should be a good deal away from the 150°C limit.
to 7) As pointed out I think it has to do with the PWM to DC conversion. Within the PWM filter they use a double RC filter each with a time constant in the order of msec. I don't understand why the time constants must be that fast. So perhaps one may increase them. The question is how to do this, I've not the abilities to work with such a micro structure.  If it's a probem of the digital resolution itself I frankly don't know, whether it can be cured at all.

[TruslowPJ]

Thanks to everyone who has posted their findings. I've ordered a DL24 armed with knowledge about it's limitations, and what needs to be done.
I ordered the version with no cooler, and will be installing an old corsair water cooler I've got on standby (just topped it up with a few extra ml of distilled water so it should last a few years). I'm planning on putting a ~12V TVS diode on the gate unless I decide to add a zener to limit gate voltage to like 7V or so, and a 15A fuse on the input so that I don't blow up anything when the stock mosfet inevitably shorts.

If the stock crappy mosfet dies, I'll probably get an IXTH64N10L2. The 60N20L2 would be a perfect fit, but it costs twice as much, I don't really need more than like 60V ever, and I'm water cooling it and don't plan to push more than 150W, so I don't know if the lower thermal impedance is really worth double the cost.

[deuteron]

Thanks to everyone who has posted their findings. I've ordered a DL24 armed with knowledge about it's limitations, and what needs to be done.
I ordered the version with no cooler, and will be installing an old corsair water cooler I've got on standby (just topped it up with a few extra ml of distilled water so it should last a few years). I'm planning on putting a ~12V TVS diode on the gate unless I decide to add a zener to limit gate voltage to like 7V or so, and a 15A fuse on the input so that I don't blow up anything when the stock mosfet inevitably shorts.

If the stock crappy mosfet dies, I'll probably get an IXTH64N10L2. The 60N20L2 would be a perfect fit, but it costs twice as much, I don't really need more than like 60V ever, and I'm water cooling it and don't plan to push more than 150W, so I don't know if the lower thermal impedance is really worth double the cost.

From the respective data sheets and your needs both Fets seem to be suitable. Because of the larger thermal resistance perhaps one should take care about a better cooling, if the IXTH64N10L2 is used. I personally prefer using Fets in the larger TO-264 package. In this case one can mount the Fet in a way that it touches the internal thermistor right away. In my measurements I compared the temperatures measured with the internal sensor and those measured with the external one, in which case I installed it right away in the mounting hole stuffed with thermal paste. Both temperature are then equal to within 1°C for most applications. As higher the load, as more the temperatures were differing. Max difference was 3-4°C at 150W. So in the future I will relay on the internal measurement (with some care at high power) so that I can use the external thermistor for other measurements.

[deuteron]

Unfortunatly our electronics engineer isn't available today. So I've to wait (hopefully only) until tomorrow, before I can remove C5 for further testing.

Anyway, I had another idea, at least for those people / applications, for which 10A max is okay. The circuit is apparently designed for a maximum current of 20A. This current initiates a voltage drop of 0.1V at the shunt resistors. The regulator (integrator) input is 100mV too to exactly counterbalance the current reading at 20A. If the load is used only with some Ampere, i.e. up to 5A in my case, the sensitivity of the regulation is somewhat coarse. So, my idea is to remove one of the shunt resistors. Then the shunt is 0.01Ohm, i.e. 100mV at 10A. In this way the resolution of the regulation may be doubled. The only problem left is then the wrong current reading of the microprocessor putting out twice the real current value. On the other hand the current can be recalibrated within the setup menu. Instead of using exactly 3A one could take 1.5A for calibration. What do you think, did I overlook something ?
Thanks,
Stefan

[deuteron]

I think I indeed made a mistake. After removing of one of the shunt resistors one has to recalibrate the unit just with the usual current setting of 3A These 3A then generate twice the voltage over the shunt. This new reading (when using indeed 3A) is then the calibration point for the microprocessor. Hopefully that's correct now.

Another question: Does somebody know, how exactly the current cailbration is done ? There is a youtube video from Indonesia, which is hard to understand. I've heard (I don't remember, Youtube ) it's more complicated than the voltage calibration, in which case you just have to connect the appropriate voltage to the load and press 'calibrate'.

[deuteron]

Hello everyone still interested. I made further measurements (scope at the gate), this time with C5 removed. To start with the result: No obvious change ! At voltage / current values such that the Mosfet has to get wide open (low voltage / high current) there are still sometimes overshootings when the current is switched on via the start stop button. It looks like, it's a bit more seldom than with C5 installed, but since this is a statistical phenomenon, it's hard to say. For the 'low current guys' of us, here comes the good news: The threshold current value, from which on the overshootings occur, is something like 8A at 3V. Below this current value the regulator works correctly. And, if the voltage is increased, its going to be better either. So, it's clear the problem comes along, when the Mosfet is opened up widely. Anyway, the  overshootings are some kind of cosmetic problem, since they only goe up to about Vcc of the integrator, I never measured more than 11V. So it's not a problem for the Fet. Depending on the model, they are able to withstand at least 20V, in most cases more. So I don't think a too high Vgs is the reason for all the killed Fets. I think it's rather the combination of a high power working point, an overshooting event and the fact that most of the used Fets are not DC-resistant. This state should be avoided by a suitable Zener in the region of 6 to 7V put between gate(+) and source(-). From my observations in normal operation Vgs never goes higher than that.