TL494 DCDC converter wide range input issues

[Zac1]

With a help of few people I managed to design fairly wide range DCDC converter using TL494

I confirmed it works in range of 30-190V DC
Has fairly stable 12V output with ~40mV peak to peak noise
Can deliver up to 3A.

However it has one fatal flaw that haunts me in my dreams. It kills mosfet Q5 and diode D2 if you suddenly apply 160V+ (high dV/dt)
It does work properly if slowly raising voltage across 40-190V. Also works up until 140V (sudden turn on).

I have managed to capture the problematic moment on thermal camera

https://imgur.com/CHKR4Ew

Basically there is a full short on Q5 and D2 for a moment, then fuse pops and both components have dead short. I have tried to remedy this situation with help of PNP clamp (Q1) but it doesn't do much. I am out of ideas how to fix this... perhaps the driver is at fault? Or maybe diode is too slow?

 

[Easy peasy]

How does Q2 work?
Why D6?

Q2 has a gate limited voltage that gives 10V out for start up ( it does rely on leakage currents to start - Vbatt is the 30-190Vin ), D6 supplies 12V from the output to turn Q2 off after start, else Q2 would overheat.

--- Updated Mar 19, 2024 ---

@Zac1 - Do you really have an 11V zener on the output ? ( from part number ) why not 13V ?

 

[Zac1]

Thank you guys for all replies so far.

Let me start with one important thing about this step down converter. Main focus is reliability.
Which means, I am very open to changing parts if there is even slight suspicion that one of them may not be sufficient.
Which is why I designed it to be able to output 3A while using max 0.3A. I want maximum headroom possible with reasonable circuit size.

So I have compiled a list of changes that would implement based on your suggestions

• D2 -> ST5L300, Schottky, 5A, 300V
• Q2 -> CPC3980, Depletion nfet, Bigger package (SOT223), a lot smaller RDSon (40ohm), 5x the current, better cooling than current BSS126
• Q5 -> IRFS4227, Bigger package, better ciss/crss ratio (less prone to self turn on), 10x lower RDSon, better power dissipation
• U3 -> ??? no candidate for now, single drivers are fairly rare, can I use in this scenario common half bridge driver?

I have on hand IR2181 which can output a lot more current than IRS21271 but I am not sure if it will work with just one mosfet, I know it should be technically possible but I have not tried it..

I did not implement soft start but that is idea worth considering.

By the way if you guys suggest some changes can i please ask for quick paint over the original circuit? It would make it easier for me to wrap my head around it and keep track of suggestions.

How does Q2 work?
Why D6?
Still miss the PCB layout..

Q2 is depletion mode nfet, which means it is initially ON without driving gate, and stays on until D6 pulls up voltage above the output of bootstrap circut. It is a nifty circuit that can generate 12V out of 15-500V with low current and turns itself off once the TL494 takes over. As for PCB layout I made sure to keep both loops very tight. I sincerely doubt it is the cause of this issue.

To perfect the simulation to match your design, will permit optimizing the design for each part, but that takes more time than a soft-start design.

I will be honest, I am not that good with ltspice. This design is a combination of many people knowledge. I tried my best to make sim better but there is always some catch.

But if you don't have 2 sq.in of Cu per watt on board for the FET and diode each, it's going to fry, even with soft-start with a 50W load.
Otherwise it could be (?)(100'C/W without.

While it is designed to output 3A, the circuit that is suppose to work with it consumes maximum of 0.3A so ~4W.

While your at it, the VFB dividers are drawing a few watts, so scale up x10R.

Good catch, will do.

@Zac1 - Do you really have an 11V zener on the output ? ( from part number ) why not 13V ?

3SMB5928B is nominal 13V zener unless im missing something?

I will compile a new schematic and will post it tomorrow.

Very very important note: DCDC dies regardless of load, even without any load whatsoever as soon as VIN is at 160V applied suddenly.

 

[D.A.(Tony)Stewart]

Transition power spikes on startup are massive so consider reducing boost cap to 22nF . Reduced RdsOn will increase Ciss and Coss so a soft start on Vref is essential.

 

[KlausST]

Q2 is depletion mode nfet

I missed that.... now it makes sense.

***

What you call "Bootstrap" indeed is the power supply for the drivers.
It may lead to problems on forum discussions, because usually Pin8 of U3. With C12 to be the bootstrap capacitor.

Also "-Batt" is misleading. Because one may assume there is a negative voltage. But indeed it is used as your system_GND (0V). Missing GND symbols make the schematic more difficult to read.

***
I just spent some time on imagining how the whole power supply and real_Bootstrap looks like at power up.
The power supply for sure has a current limit on power up.
C12 relies on empty output capacitors and can only be charged with (negative) current through L1.

There will be a time where the ouput voltage is 6V.
This is not high enough to supply your_bootstrap_node via D6 ... while it is also not low enough to charge C12 to a proper level.
Not high enough to work peoperly and not low enough to work properly .... means: what happens then?

So maybe you are doing several tests .. with increasing input voltage. But the fault you see is not caused by the "increasing input voltage" but indeed the root cause may be the only partially discharged output voltage. As said: maybe.

***
For the third time: PCB layout!
You want to focus on reliability. In many discussions here we see that the PCB layout has big impact on reliability. Causing unstable situations, causing voltage spikes due to stray inductance, ground bounce due to currents ...
--> there is now way to talk about reliability .. without being able to check the PCB layout.

I will not come back until we see the PCB layout.

Klaus

 

[Easy peasy]

you really need to test the start / bootstrap ckt as prev described, under hard start - hopefully it is 190V proof (?)

next likely is the ( unseen by you ) surge current at start due to no soft start or pwm limit - at hard start - heating the diode, and then the fet and then - poof - this gets worse for higher Vin step

What is also unseen by you is any ring up on the Vin when suddenly applied - this could be a slightly wonky psu that goes up a bit in response to the sudden load, or an LC ring up if supplied by a real battery ( we can't see the full real ckt )

matters are probably exacerbated by the unknown quality of the gate drive - 200mA is too little drive - and we don't know if it holds the gate off properly either, as the source falls rapidly

the gate drive IC may be misbehaving for > 140 Vin step applied, due to dV/dt

a lot of little and not so little things can come together to make a ckt go bang for > 140VDC applied suddenly

EP

--- Updated Mar 19, 2024 ---

" C12 relies on empty output capacitors and can only be charged with (negative) current through L1. "

the current is not negative.

also - what does this mean ?
" The power supply for sure has a current limit on power up. "

--- Updated Mar 19, 2024 ---

" While you're at it, the VFB dividers are drawing a few watts, so scale up x10R. "

is this true ? 12^2 / 10k = 144/ 10,000 = 14.4 mW (?)

 

[Easy peasy]

Klaus may well be correct about the inability of the boot strap psu to deliver up to 12V quickly enough - due to the resistance of the very small depletion fet - and its undoubted extra heating as it has to block more and more volts, causing R-ds-on to go even higher than 700 ohms - thus the main fet will not get sufficient drive voltage on its gate and will heat during start up - this can easily be tested with a scope on Vout and the output of the boot-strap start ckt at power up.

 

[Zac1]

Klaus may well be correct about the inability of the boot strap psu to deliver up to 12V quickly enough

That was my original thought about the source of issue.

Thus I have compiled a list of new changes and new schematic

D2 -> ST5L300, Schottky, 5A, 300V
Q2 -> CPC3980, Depletion nfet, Bigger package (SOT223), a lot smaller RDSon (40ohm), 5x the current, better cooling than current BSS126
Q5 -> IRFS4227, Bigger package, better ciss/crss ratio (less prone to self turn on), 10x lower RDSon, better power dissipation
U3 -> IR2181, half bridge used only with high side, 10x the power, 1.9A
C12 -> 100nF, halved the boostrap cap capacity
R20, R21 -> scaled VFB caps 10x
NEW: C20, R22, R23 -> Soft start circuit calculated for 48%, and ~100 cycles
R17 -> 47 ohm, increased current limit of startup circuit twice (not sure about this yet, will try to scope current one)
R18 -> 9.1k, increased slightly voltage for startup circuit
Q1 -> MMBTA92, just in case, capable of withstanding higher voltage

Original V1.5 layout, yes i am aware it can be done better but i have limited space and have to work around gigantic caps on the board on the other side:

 

[Easy peasy]

Is there an LC filter supplying this board ? i.e. between the batt and a the board?

p.s. the volt sense resistors were fine as they were, < 15mW ( see above )

 

[Zac1]

Is there an LC filter supplying this board ? i.e. between the batt and a the board?

No, its directly connected to battery.

How about the changes i made? Looking good? Any suggestions?

 

[Easy peasy]

time will tell

 

[Zac1]

Finding appropriate Schottky diode has proven to not be trivial. All high voltage schottkys have relatively high Vf which is far from ideal but I came up with 3 candidates:
CI04S65E3, IV1D06006P3, STPSC4H065B-TR

Schematic remained pretty much the same with exception of D2 which above candidates i have in cart.

when it comes to layout, here is what i cooked up, suggestions appreciated:

Hot loop red arrow

3D view:

 

[KlausST]

Hi,

We need to get a step forward. To locate the problem.

Try this:
Use a seperate power supply of about 12V and power up your STARTUP node.
Now the ICs should be properly powered up .. and running, even the bootstrap circuit.

Now do different tests with powering up +BATT node.
The resupts show you if the problem is caused by IC startup or by the power stage around Q5.

***
I still think thea STARTUP is not a good description for this node. It is not used only during strtup but all the time. It simply is a power supply.

***
and yes, I agree that the PCB layout is ot optimal. I miss a clear solid GND plane with all components on the opposite side. You say lack of space, but there still is a lot of space on the TOP side ... but you use it to "sew" the GND fragments on TOP and BOTTOM side together. ... Which will never be as good as a solid GND plane on a SINGLE side.

Next important is a clean, straight (or low area) power path: Input --> capacitors --> Q5 --> D2 --> L1 --> capacitors --> output. (When Q5 is ON)
but also: GND --> D2 --> L1 --> capacitors --> GND (when Q5 is OFF)

Next is Q5_Gate path ....

Klaus

 

[Zac1]

We need to get a step forward. To locate the problem.

Yes I would love to do so however there is one big problem.

I do not have power supply capable of outputting 160V+ with current limit which means this test would be one-time only. And this is last mosfet i have so i would prefer to do it non destructive way.

What I can do is to supply startup circuit with 12V and power whole thing with bench PSU (up to 60V) which does have current limit (but then again it doesn't trigger destructive effect)

lot of space on the TOP side

There is a bit of space but not a lot, capacitors are not movable, top and bottom is occupied by massive copper busbars.

I am already bending rules a bit by making cutout in bottom busbar, may need to stack it.
Those paths are suppose to carry over up to 150A

 

[KlausST]

There is a bit of space but not a lot, capacitors are not movable, top and bottom is occupied by massive copper busbars.

If this is the TOP side, then you previously showed us the PCB upside down.
And those capacitor are nowhere to find in your schematics! (Or I´m blind)

Again: many pieces of GND do not give the same quality as one solid layer.

150A??? How? In post#1 you talked about 3A. Why now 150A???

Klaus

 

[Zac1]

If this is the TOP side, then you previously showed us the PCB upside down.

Yes, sorry about confusion, they are above schematic as i didn't think they were relevant to PSU. They are not soldered on test boards anyways. But I do have to work around them.

This is how ground plane on bottom side looks like:

And this is top side ground plane. I think it is pretty solid overall (but I could be wrong).

 

[KlausST]

And this is top side ground plane. I think it is pretty solid overall (but I could be wrong).

Solid means solid. A solid rectangle. A solid sheet of copper. No additional traces, no splits, no empty spaces...
Your "plane" is no plane. They are connected fragments. Don´t get me wrong: I don´t want to annoy you and for sure the big copper areas bring some benefit against "simple traces" ... but not that much benefit as you might think. Neither for DC currents nor for HF.
95% of your (stiching) vias are in areas where there will be no current flow. No current flow --> useless.
--> Place your vias where they are needed, where you expect current flow.

For "not that experienced PCB designers" the easiest to follow recommendation for a good layout is to use one completely solid layer.

Later ... The more experience you gain, the more you understand where will be current flow ... Thus with more experience you may be able to "cut your GND_Plane" into pieces without losing performance.

Your layout:
The hole in GND copper (blue) around Q5 acts as an antenna. the bigger the area the lower the frequency and the more prone to pick up frequencies.
Then it is enclosed by a copper loop, creating a high Q resonance. No matter how many vias you are using, as long as the hole is "open" you get this resonance problem.
So regarding this antenna ond resonance it would be better to leave the copper ring around "unclosed". But you should have some experience to find out where´s the best location to "open the loop".
Additionally your power current flow

My opinion with your GND_plane may sound harsh ... but it´s not meant offending. It is meant to be focussed on the application´s failure problem.
Until now we don´t hve enough information to be sure what exactly causes the problem ... but the PCB layout MAY be one of your problems.
And I try to sensitize you on this problem in a way that your future projects start with a true solid GND plane ... to be sure to exclude this problem from the beginning.
--> One problem less to cure.

Klaus

 

[D.A.(Tony)Stewart]

" While you're at it, the VFB dividers are drawing a few watts, so scale up x10R. "​

​

is this true ? 12^2 / 10k = 144/ 10,000 = 14.4 mW (?)​

No

I saw 160 V on the result of modifying his LTSPICE file and measured power (Alt+Click) without thinking/knowing yet what were the design specs for the output.
It's should be 12V.

Then I thought why bother stressing a high side switch when a low side switch could be ramped up much easier. Ground is wherever you choose as 0V.

 

[Zac1]

but not that much benefit as you might think.

They are mostly to help the current flow from/to big caps.

Place your vias where they are needed, where you expect current flow.

maybe this view can convince you:

One thing that may not be obvious from this design is that vast majority of current will flow from/to bulk capacitors, reaching 150A (up to 168V)
DCDC is just a tiny droplet that is meant to sip a bit of power (up to lets say 5W), its stability isn't even that important as making sure Phase mosfets gets full power.

Unfortunately due to size/design constraints I cannot design perfect PCB, I have to work with what i got. And i need to make the "cutout" on the top due to screws/busbars.

The hole in GND copper (blue) around Q5 acts as an antenna.

Which hole are we talking about?

My opinion with your GND_plane may sound harsh ... but it´s not meant offending. It is meant to be focussed on the application´s failure problem.
Until now we don´t hve enough information to be sure what exactly causes the problem

That is fine, I'm not offended, I know this design is subpar, if I had some tested solution I would take it but so far after couple of months of research i have yet to find something that fits design constraints and I will try whatever I can.

So far I have tested 3 "floaty" designs (unstable), flyback (way too big) and this "classic" DCDC which so far works the best.

I guess i should just order PCB + components again, try to record what happens on boot, and if problem persists come back and try more troubleshooting.
Hoping new ones won't burn as fast.

 

[D.A.(Tony)Stewart]

I would work on your simulation of diode and FET Cdg capacitance with better parts and simulate the failures.

Ctrl+ click for power plots on selected part after adding a plot plane then select it.

 

[Zac1]

I would work on your simulation of diode and FET Cdg capacitance with better parts and simulate the failures.

Ctrl+ click for power plots on selected part after adding a plot plane then select it.

I wasn't even able to get this simulation fully working, so unfortunately its beyond my ltspice skills. It works 1/10 times only when i mess up with values in some weird ways.