300A bldc esc, mosfets ringing, help

[Mihai_F]

Hello community,
i have been working on a big project for the past 6 years , https://www.avrfreaks.net/forum/electric-propulsion-system-ppg-and-ultralights, you can find it here for details, i can post it here also if there is interest.
The main problem that i have is that low side FET-s (IRF150P220) are ringing at turn off, the gate driver are IR2110 with isolated supply on high side. I tried with 15 ohm and 100 ohm gate resistors but the ringing is still there. I attached 2 scope shots with both gate resistor versions an a few photos of the arrangement of the FET-s, the copper on the power board is 1 mm thick. This is the third version of the board (v1.3) and this is the best layout i can do, considering that is in a box and v1.2 flew (check the link). v1.2 was with IRFP4468 and did not have nowhere near such ugly ringing.

 

[Jrbe]

My only add is that I have not had good luck with using picoscope products. I was not using the one I had on anything this sensitive and I knew it couldn't be trusted.
It might be accurate or might not be. Basically worth seeing if you can borrow a known good scope before chasing a ghost.

 

[mxlemming]

With the 100 ohm gate resistor you're hitting an oscillation as the switching occurs and the gate drive goes through the MOSFET Miller plateau. With the 15 ohm resistor you're getting ringing from inductive artifacts, but it's quite likely as jrbe says that this is probe artifact not actual ringing.

If it is actual ringing, you'll probably want to snub it rather than slow the switching. Energy lost to avalanche on ringing vs energy lost to slow switching... The slow switching will be more energy.

Low gate resistance is better in this case. 100 ohm is way too high.

Look at the double pulse tests in my MESC thread, i cover a whole load of probing stuff I learnt myself along the way. You'll want a ground spring and a proper probe if you haven't got it already.

Also look at user Peters' thread on his controller... Similar to yours in terms of MOSFET type and talk care taken and detail on the testing.

 

[Mihai_F]

Well i know that the measurements are not 100% acurate, but v1.2 with same layout and IRFP4468 fets worked with 200 amps at 84v, and now with IRF150P220 fets the ringing is way bigger, and i can not go past 50 amps becouse fets blow up, and yes they did blow up, all 18, preaty expensive. I think the ringing is way bigger becouse irfp4468 had only 1nf Cds, and irf150p220 has 3nf Cds, and that excites things higher. I also know there is ringing because i get preaty high EMI and disrupts uart comms with display module, v1.2 did not do that. This https://www.youtube.com/watch?v=NTU_feiQHvM&t=1s is 2 minutes out of a 30minute test, i mean i'm baffled that changeing the fets with "better ones" me thinks, is actually worst. The change was made, because i increased battery voltage from 20s to 22s (84v to 92,4v), and fets with higher Vds were needed, plus 4468 has a low Ciss/ Crs ratio and were harder to drive (lots of gate and turn on ringing) and becouse added capacitor (33nf)to GD go get better ratio. And i also have a different question when ceramic caps are overworked hard how do they fail?

 

[mxlemming]

The main problem is that your MOSFET is imaginary. It doesn't exist.

Your second problem is that your layout is hideous. Both aesthetically and electrically.

Your third problem is that despite some good advice you seem to be avoiding actually reading the threads I listed. Many people here have made 300A controllers, the answers to how to do it are on this forum.

Ceramic caps often fail short, the ceramic dielectric punctures and then they burn. Or just short whatever they're meant to be buffering.

https://endless-sphere.com/forums/viewtopic.php?f=30&t=110674&p=1654507&hilit=Peters+controller#p1654507

https://endless-sphere.com/forums/viewtopic.php?f=30&t=107672&hilit=MESC&start=100

https://endless-sphere.com/forums/viewtopic.php?f=30&t=109999&hilit=Thunderfoc

Zombies is the master of the to247 MOSFETs with non exploding ness, he's done 600A+.

https://endless-sphere.com/forums/viewtopic.php?f=30&t=106688&start=50

 

[Mihai_F]

Let's clear a few things first:
1. Just now i realized that i made a mistake on typing the FET-s part number, it is IRF150P220, sorry for that mistake, i will edit the post. In the avrfreaks link that i shared in the first post, there is the corect part.
2. I did read all the post you mentioned, very valuable information i have found here, and also while making my esc i did go thru some problems like they have encounterd.
3. Did you even look at the links i sheared, i mean this project did not start yesterday, plus it worked until changing fets.
4. Aesteticly hideous, sure, it can be ugly to some people, and nice to oather, that is just subjective.
5. Electrically hideous, yes it is compared to your design or oatheres with fancy multilayer PCB-s that have power traces and TOLL fets with inductance down to single digit nH, that is the ideal way to do things like this ESC-s, but it is not the only way to do it, as i have seen around here. You can switch 100A in 50ns thru 2nH , or in 500ns thru 20nH, with some minor power loss penalties. It all depends on what technologies you have available, in my country it is difficult to acces technology like that (multilayer pcb-s and smd fet power boards).
I have seen here ESC-s with PWM at 200Khz, that alone will cause BLDC motor iron heating and losses (not even mentioning about copper skin efect) way way higher than the gains of FET-s switching on and off in 50ns, this is just a side note, anyway.

One thing that is not to be desregarded in my design is its thermal performance, verry good heatsinking, thick cables (35sqmm), it could run WOT at 200A all day long and nothing heats up over 40 deg C in 25 deg C ambient. This esc is intended for aircraft usege, where thermal stability over long periods of high power is a must. On an high power e-bike you will not go WOT for more than a few seconds unless you go on highway long distances.

Anyway, i will carry-on with valuable info found here, i was hoping to find some opinions why different FET-s make different turnoff ringing, but i think i found some answers in some of peters esc post-s (different Cds gives different turnoff ringing) .

 

[zombiess]

Sorry if this comes off a bit harsh, but you are venturing into aerospace engineering in what appears to be an unprepared manner. The following isn't meant to be negative, it's meant to make you think.

Just my personal opinion, but I think flying with the controller setup pictured is bad risk management.

There does not appear to be any fault management. Do you at least have a battery disconnect system? Why is fault management so important? It's a bad day when you are in the air wishing you were back on the ground. It's an even worse day (probably shorter too) when you are in the air, on fire, wishing you were back on the ground. Controllers fail and can plasma ball, MOSFETs almost always fail short, batteries catch on fire, etc. It's good practice to address these types of situations before flying. I've seen several videos of battery fires caused by failed controllers, they tend to go bad really fast.

General comments on the power stage:
Your gate drive traces should never cross into your power pass section and should make a kelvin type connection to minimize noise and ground bounce out of sensitive areas.
Gate drive is not using dedicated isolation or drivers with fault handling.
The DC bus does not appear to be overlapping <--- this is probably a big contributor to your ringing
Double layer PCBs aren't good enough for this application, you don't get close to the same noise immunity you do from +4 layers.
Do you have enough DC Link capacitance to handle the ripple current?
Airplanes tend to be vibration prone, is everything secure?
Has all internal wiring and all components been isolated from abrasion?

I encourage you to keep designing, but not flying with your controller.

 

[mxlemming]

I did read your thread on avr forum. It doesn't seem like you got a whole load of power electronics advice there but... They did point out that running without a fuse with a1200A battery might not be a great idea.

I also note that you have no obvious current sensors. Where's the feedback to stop it cooking itself? It appears to rely only on resistance and gentleness with the throttle.

My point in the above post is basically that if your controller was built with as much care as your typing, then I mirror zombiess recommendation - please don't fly with this. At least it won't fall out of the sky being a hang glider but it might catch fire and a fire a few hundred meters in the air doesn't sound like fun.

TI and on semi make some very nice isolated gate drivers. Broadcom and on semi make nice ones with desaturation protection.

You should probably convert your power stage to a 4 layer PCB with the power layers overlapping and try to minimize the inductances - your MOSFETs probably have like 20nH but your layout seems to have a loop between+and ground of several centimeters diameter.

The ringing from Cds isn't likely to kill it, and doesn't scale with current. What really kills MOS is over voltage, parasitic turn on, shoot through... These are high energy events. The energy from charging and discharging 2nF is pretty small.

I also can't see much decoupling. I see the little red capacitors but frankly their package inductance is huge and they're not eliminating the huge loop. At 1uF each I'd say they're inadequate for your switching current and speed.

You might have got away with the old MOS because they switch slow and so parasitics and gate bounce take longer to propagate. You want to use the latest fast switching Infineon FET? Really need to have a fast switching layout.

Nothing wrong with adding copper busbar to a4 layer jlcpcb. Get your low inductance and high current.

 

[Mihai_F]

For fault management, i have emurgency battery disconect, the controller is in a 1,5mm thick aluminum enclosure so plasma and fire can not get out of it (i tested it), and if FET-s go shorted the battery has enough energy (4,88Kwh 1200A continous) to vaporize them and anything in the controller that stands in its way (also tested), the "advantege" of leaded FET-s are like fuses and they're legs go pop when the FET-s get sorted, i know this is kinda brute force failsafe but it worked. The only thing that worries me is battery fire if something mecanicly shorts the power cables or puncures the battery, witch also is in 1,5mm thick aluminium casing, but that is not fire proof, just small impact proof. All the wires secured, command board is soft mounted inside, so vibration can not pass to it, the entire controller unit is mounted with vibation isolator to the airframe and electrically isolated.
Yes i know this controller is way far from beeing fault proof, there is lots of room to improve and it will. For now i want to make it work again (with irfp4468 fets worked fine) , and then reliability comes thru testing, if works and does not fail.
I have some picture from someone who smoked FET-s on an MGM - compro ESC, a 500A version for aircraft, witch is considered tip of the sword in ESC industry, big players like Airbus and others collaborate with them (MGM). That ESC has everything you can think of in terms of technology multilayer boards, isolation on gate drives, multiple failsafes and all the whistles and bells. So if that ESC did not keep the smoke in its FET-s, then what will?
Everything in aviation is asumed risk. The aircraft in witch a took my pilot license, has in ICE rotax 912, and in there's manual they say the engine can stop at any moment from unknown reasons, and this is a well respeced engine in ultralight industry. That aircraft with that engine flew 350 hours just fine with the choke cable rubbing the engine cowling everybody said at inspections it is fine, and one day at around 350h, in flight the engine decided to shake very violently, and a very quick emurgency landing was carried out. The problem / symptom was intermitent and after a week of trile and erorr, mechanics found that the rubbing choke cable induced vibration in left carburator and that made the fuel level in the bowl to rise up making a rich mixture and that desincronized it (carb) from the right one, hence the engine shakiness (dual carb engines must work in sinc). Now why only after 350hours that started to happen, no one knows, time will tell.
This is ultralight aviation witch is uncertified, hence unknown resons of engine failiure and has asumned risks. I fly with the mindset that engine / motor can quit in any moment and at worst moment posible, (like in the takeoff la 50m AGL).
I understand your worries, i have mine to, but at the end of the day, i hope i have enogh wisdom to know when to fly and when to not with those risks (manny or few) asummed.

 

[Mihai_F]

It does have hall current sensor on the positive bus.
I have to wap my head aroud and figure how to make a 4 layer power board, i was to fixated on 2 layers.
Those small 1uf caps, failed after a while (melted) ESC was fine, then switched to ceramics, haf 4 banks of 5x1uf in parallel, they failed too, and took with them a trece and all 18 fets as a result.... (first failure after a while of working)

 

[mxlemming]

For this current and the stay inductance in your layout I think you probably want an order of magnitude more decoupling capacitance, and critically, they'll be taking the pulses of max current at switching so more of them needed to take that.

I somewhat agree on the expensive complicated controllers fry front... My strategy has been to reduce the complexity as far as I can without removing functionality, and it's served me well, but that's using the ultra low inductance FETs and careful power stage.

However, I've struggled with currents over 200A with my simple designs, which i think is where zombiess and peters' techniques with the isolation and desaturation come into their own. It is possible though, the VESC 75/300 and even UBox seem stable at 300A (though they will eventually fry from heat) and they have simplistic gate drivers with just excellent layout. They do go bang a lot though...

 

[Mihai_F]

I'm thinking to use polypropylene caps (i have some 1uf 250vac big ones) for decoupling or ceramics, but i'm afraid of those becose they fail with no worning and are kinda distructive, aldo pp caps are not as good as ceramics at decoupling, il try it and see.

 

[zombiess]

I'm thinking to use polypropylene caps (i have some 1uf 250vac big ones) for decoupling or ceramics, but i'm afraid of those becose they fail with no worning and are kinda distructive, aldo pp caps are not as good as ceramics at decoupling, il try it and see.

Polypropylene caps are often the best choice for DC link for a high reliability design. They provide high current capability in a volume smaller than most electrolytic solutions with a longer working life. They are also used for decoupling and I generally would use them vs a ceramic capacitor depending on the application. MLCC ceramic capacitors are going to suffer from DC bias effect, so if you need a specific value, you'll need to over spec the amount of capacitance you need along with obtaining higher working voltage devices. Another downside with ceramic caps is that they are fragile. The larger the package, the more prone it is to physical stress which can be caused by thermal gradients and cycling. When making a decision on what type of capacitor to use you need to take both electrical and mechanical considerations into account.

 

[mxlemming]

The big issue with polypropylene caps is that by their shear size, they have big inductances and therefore really don't decouple in the same way ceramics do.

They're probably fine as long as you're switching slowly enough that their inductance doesn't limit their usefulness but just consider... They're as big or bigger than the MOS you're decoupling.

Ceramics really do have issues with fragility, see the board Thor made where one day his special design went pop pop pop.

I'd err on the side of ceramics but for this high reliability application take very special care to mount them properly with the thermal strain reliefs around their pads and away from mounting screws. No soldering then to busbars.

 

[zombiess]

I'm curious as to what benefit you are getting adding decoupling capacitors. I've found that with a good layout that my DC Link cap is enough and no additional capacitance is required unless I add a RC/RCD snubber, but then the application is different. I've also found adding capacitance down stream of the DC Link caps can cause additional ringing to show up.

The big issue with polypropylene caps is that by their shear size, they have big inductances and therefore really don't decouple in the same way ceramics do.

They have big inductance vs a MLCC, but even packages with 40mm wide pins are only about 20nH inductance but I don't see an issue.

They are also one of the best choices to meet size constraints. I use two 90uF caps in one of my designs and each one is rated to handle 35A RMS of ripple current. That's 35A in a package of 35x50x42mm which is tiny for its capability. It's also only 14nH of inductance, 1.8mOhm ESR and it only costs $10 and will out perform most any other solution in pretty much every category.

When you start designing for high performance and higher power you need to step away from the thinking used in lower power. Often times the trade off of additional cost and size to do it properly is easily justified by life safety requirements or asset value. Doing it correctly is often the lowest cost option when you look at the big picture.

 

[Mihai_F]

When testing the v1.2 with 4468 fets i had 6 nichicon 1000uf caps, witch had only about 11A ripple and about 6mohm esr total, and after 2 30min test when power was cycled repeatedly from 10 to 16Kw, the caps did not even get worm, now they had a lot of cooling (prop wash).
V1.3 has 6 chemicon KZN series caps 1000uf, total 24A ripple and 2.8mohm esr, i think they will survive even better, but testing will confirm that.
I also considered using pp dc link caps in a future version, but i have package size restrictions, i have to find someting narow and flat domething like 18x55x40mm 2 pieces max, no space for more.
I have read Thor's controller post when the row of ceramic caps went pop, that is what happend to me, an entire row soldeted to the busbar wet pop and then mayhem of the fets, in that test day was 6 or 7 deg C out, so the busbar was shrinked by temp and the ceramic caps were mechanicaly stressed, they lasted 6 minutes... Now i have an ideea of putting the ceramic caps (with 1mm between them) on a thin and long (9cm) pcb strip (about 20 pieces on each side), and then attach the strip with multiple short wires to both bus bars, so then the thermal expansion of the power board does no transfer to the ceramic caps, plus some big 1uf 250vac pp caps,

 

[mxlemming]

I'm curious as to what benefit you are getting adding decoupling capacitors. I've found that with a good layout that my DC Link cap is enough and no additional capacitance is required unless I add a RC/RCD snubber, but then the application is different. I've also found adding capacitance down stream of the DC Link caps can cause additional ringing to show up.

The big issue with polypropylene caps is that by their shear size, they have big inductances and therefore really don't decouple in the same way ceramics do.

They have big inductance vs a MLCC, but even packages with 40mm wide pins are only about 20nH inductance but I don't see an issue.

They are also one of the best choices to meet size constraints. I use two 90uF caps in one of my designs and each one is rated to handle 35A RMS of ripple current. That's 35A in a package of 35x50x42mm which is tiny for its capability. It's also only 14nH of inductance, 1.8mOhm ESR and it only costs $10 and will out perform most any other solution in pretty much every category.

When you start designing for high performance and higher power you need to step away from the thinking used in lower power. Often times the trade off of additional cost and size to do it properly is easily justified by life safety requirements or asset value. Doing it correctly is often the lowest cost option when you look at the big picture.

The ceramics next to the FETs decouple them, otherwise your switching path for x hundred amps at 20kHz includes the loop back to the bus caps. Apply V=Ldi/dt and use dt is 100ns, L is 20nH, di is 200A and you get a spike of 20x200/100 is 40V spikes. If you choose to keep that extra 20nH from the polypropylene cap out a long busbar, you have to switch slower, use higher voltage components or add big snubbers which effectively become the decoupling caps.

You can't cheat physics.

I get that at some point your FETs become so big that little SMT stuff and ceramics aren't viable, but then you just have to accept you eat the losses of slow switching. But that's probably not so bad because big motors tend to be high inductance and will accept slow switching.

The turnigy ca120 150kV on my desk that I'm trying to drive? That's 1.5uH and 4 milliohm phase resistance. You can't switch slowly for that. That's pretty much a dead short with 6awg wire.

 

[Mihai_F]

The v1.2 had about 500ns turn off and this on the output at 75V and 130 A

I think i'l stick to that speed because it worked fine in terms of overshoot and heating of the FET-s, the low freq wave was probably ringing between DC link caps and decoupling caps
I'l next week try snubbers between switching node an GND, and some decoupling caps, this week i had a lot of reading on this forum on others projects and now i have some ideas to try to make it (v1.3) work again.
PS it amazes me how you pass 200A on such thin (AWG6) wires, they must have some losses, I mean i have 35sqmm (AWG1.5)....

 

[mxlemming]

The v1.2 had about 500ns turn off and this on the output at 75V and 130 A
gate and fet1.png
I think i'l stick to that speed because it worked fine in terms of overshoot and heating of the FET-s, the low freq wave was probably ringing between DC link caps and decoupling caps
I'l next week try snubbers between switching node an GND, and some decoupling caps, this week i had a lot of reading on this forum on others projects and now i have some ideas to try to make it (v1.3) work again.
PS it amazes me how you pass 200A on such thin (AWG6) wires, they must have some losses, I mean i have 35sqmm (AWG1.5)....

It might be that you are just better off sticking with the older, (probably cheaper) FETs if that's what worked. As a general rule, FETs with larger reverse transfer capacitance like the IRFP4468 switch slower because of the feedback from the switchnode through said capacitance. To achieve the same slow switching from a modern high speed FET would be a lot trickier, since they have very low reverse transfer capacitance and are intended to be used in a different way. I am not even sure it is possible to make them switch slowly practically, you just have to run with the new fangled way of laying everything out with ultra low inductances.

I rarely use 200A, I think I only ever saw that much for double pulse testing on my small controller, and the big VESC based one I have is normally wired up with 8 or 10awg, but at those currents, everything gets hot pretty fast and you soon end up at silly speeds so it is short term only.

 

[JackFlorey]

The big issue with polypropylene caps is that by their shear size, they have big inductances and therefore really don't decouple in the same way ceramics do.

Yep. When you look at their impedance over frequency, their lowest impedance happens at a fairly low frequency due to that effect.

For critical filtering applications often you have to use a layered approach. Small SMT (preferably COG or NPO, but X5R can work too) for high frequency. Polypro for medium frequency, and bulk (electrolytic most often) for low frequency. And watch the inductances in the large caps - you can easily end up with a resonance that will look to you like ringing.

Also it's worth noting that your first line of defense should be your PWB. Put traces that need decoupling on top of each other with as much copper as possible; that's going to give you a very high Q cap. (And of course, if you do NOT want them to couple, route them perpendicular to each other with a ground plane between them if possible.) This means at least a 3 layer board (4 layers is standard) but you can reduce expenses by making the external layers much thicker (1 or 2 ounce) with inner layers being cheaper half ounce.

 

[Mihai_F]

I tried this morning with snubbers (4.7n+5r) on low side fet-s and 47R gate resistor, things improved big time at 10A load (compared with the first measurement in the post with no snubbers), now in the next days i'l go out put the prop on and give it load and see how it turns up, hopefully not

 

[Mihai_F]

This last measurement was made directly on FET legs close to case with tip of the probe and the short coiled gnd aroud the tip, that is not easy to make and grate care must be taken not to short circuit FET legs to the busbars or to themselves, that will be even more difficult in the open air while prop blast 60km/h wind and 90kgf becouse everything is shaking and moving around, so i think i will solder a short 5cm piece of RF 50ohm cable to the FET legs and connect my probe to that cable in a low inductance fashion, hopefully that will keep measuring attracts to minimum. Now i did a few measurements with my picoscope to some know waveforms to validate that measurement artifacts are not to exaggerated, one of them was the switching node of the LM5017 on the command board, witch showed only about 4V overshoot and undershoot and no ringing witch seamed reasonable acureate (with low inductance gnd on probe tip).

 

[mxlemming]

This last measurement was made directly on FET legs close to case with tip of the probe and the short coiled gnd aroud the tip, that is not easy to make and grate care must be taken not to short circuit FET legs to the busbars or to themselves, that will be even more difficult in the open air while prop blast 60km/h wind and 90kgf becouse everything is shaking and moving around, so i think i will solder a short 5cm piece of RF 50ohm cable to the FET legs and connect my probe to that cable in a low inductance fashion, hopefully that will keep measuring attracts to minimum. Now i did a few measurements with my picoscope to some know waveforms to validate that measurement artifacts are not to exaggerated, one of them was the switching node of the LM5017 on the command board, witch showed only about 4V overshoot and undershoot and no ringing witch seamed reasonable acureate (with low inductance gnd on probe tip).

I did very careful measurements of lm5017 and posted the results in peters' thread for comparison. There is virtually no ringing on it atall.

You've changed the gate resistor value as well. You haven't just hit a sweet spot in the switching speed have you?

Generally at higher currents your snubbers will need to be bigger. 10A and 5R is about 50V, which is sensible in the context of your bus voltage.

At 300A that 5R isn't going to do so much... Your time constant on that pair is 25ns which isn't far off considering the ringing frequency, but don't be surprised if at 300A you need much bigger parts - like 47nF and 0.5 ohm and then they need to be rated for a few W

It's quite likely though that i have a bit of a misunderstanding on this though since peters managed ok with just tiny snubbers. The thing that surprises me is that the difference between the energy in his oscillations (1/2i^2xL with 20nH 200A) and in the capacitors of the snubbers (1/2CxV^2 at 3.3nF ~80V) is quite different. I get how it damps but not how it can snub in a single rise time.

https://endless-sphere.com/forums/viewtopic.php?f=30&t=110674&start=25

Anyway, it seemed to work so yours might well work well as well.

 

[Mihai_F]

I did very careful measurements of lm5017 and posted the results in peters' thread for comparison. There is virtually no ringing on it atall.

You've changed the gate resistor value as well. You haven't just hit a sweet spot in the switching speed have you?

Generally at higher currents your snubbers will need to be bigger. 10A and 5R is about 50V, which is sensible in the context of your bus voltage.

At 300A that 5R isn't going to do so much... Your time constant on that pair is 25ns which isn't far off considering the ringing frequency, but don't be surprised if at 300A you need much bigger parts - like 47nF and 0.5 ohm and then they need to be rated for a few W

It's quite likely though that i have a bit of a misunderstanding on this though since peters managed ok with just tiny snubbers. The thing that surprises me is that the difference between the energy in his oscillations (1/2i^2xL with 20nH 200A) and in the capacitors of the snubbers (1/2CxV^2 at 3.3nF ~80V) is quite different. I get how it damps but not how it can snub in a single rise time.

https://endless-sphere.com/forums/viewtopic.php?f=30&t=110674&start=25

Anyway, it seemed to work so yours might well work well as well.

I saw your post on testing LM5017, from there (peters thread) i got inspired, and did my test on LM5017, the layout is nice and tight, so my 4V overshoot and undershoot might be measurement artifacts.
The gate resistor, i tryed a few values, 15R was fast and nice but ringing was high, 100R (as you pointed out was way to slow) and ringing was the worst plus some bad oscillations in the miller plateou, 47R gave a rise/fall time about where i was aiming (~500ns) becouse that created a workable overshoot in v1.2.
About snubbers, as i understand from peters thred(10kw controller...) at turnoff the rise of Vds excites the LC formed by the fets Coss and fets legs inductane, and that rise starts the oscillation, hartley oscilator, like thor mentioned here https://endless-sphere.com/forums/viewtopic.php?f=30&t=109999&hilit=Thunderfoc&start=100
, so the snubber snubbs that oscilation. At low current inductive spike is small 40nH 500ns 10A gives 0.8V and that is blanketed buy the ringing, but a at 300A gives 24v, witch is why i think snubbers work, if they can snubb 40v ringing, they can snubb 24v inductive spike. And i believe that is why my fets died, 90v bus plus 40v ringing plus 24v spike easily got over max Vds 150v.

 

[mxlemming]

Yes, this makes perfect sense when you're using slower switching to tame the inductive spikes and the snubbers are only there to tame the mosfet induced oscillation.

I keep coming back to thinking about my circuits where I bludgeon the gates with a strong enough drive current that they turn on in 1/10th (50ns) the time and basically can't oscillate.