Maximizing Heat Dissipation in High-Power SMPS Design

[Salvador]

oh thanks for showing me this site , looks really good, I even checked up and started to order to see the shipping prices and they are reasonable atleast compared to what I saw elsewere, the problem is that I cannot finish the deal because it seems they are still developing their business and the homepage is temporarily closed so i cannot proceed beyind the input your data stage for odering , maybe will call them tomorrow to see what can be done.

 

[Salvador]

so i got the fch104n60 from the RS company 9by the way thanks for telling me about them)
Now I wonder what kind of a startup load i could use i mena the minimum for a test?
i kinda don't feel like putting the 500w bulb which had 10 ohms when cold. since the use for the smps will be in an amplifier it would rarely see such low resistance (typical speaker load would be 4 to 8 ohms but that's an impedance load not a purely resistive one and one rarely plays speakers to max)

i have a 230w 15 ohm bulb , i don't know what else I could use to see the output power of whatever i have built there?
or maybe i should take the 500w bulb and just use a reostat to slowly bring it to full power.
also I wonder does the soft start capacitor on the sg3525 acts only as a time delay depending on the capacitance used or does bigger capacitance actually startup the mosfet driving IC's slower? like more gradually.

 

[Baluncore]

Take the output and connect across it several filament globe sockets with switches. That enables you to switch in a variable load during your tests. You should select and design your load by using the cold resistance of the filaments.

Your +/–70V represents a 140V DC output. That is insufficient to fully light a 230V filament lamp. The cold resistance of a globe will be about 1/10th that of a bright globe. We know that because resistance is proportional to absolute temperature. Room temperature is 300K while an incandescent filament will be closer to 3000K.

I have 240VAC and use an equation to estimate Rcold = 3600 / Filament wattage. So a 100W filament globe will have a cold resistance of 36R. 75W will have 48R. Each switched filament will be a current load. The maximum current will be less than Vout / Rcold since the filaments will be warm, radiating IR. It will not be the cold or the hot resistance, it will be something in between. You can always measure the current and voltage to determine the filament temperature.

You can buy 230V filament globes from 15W to 400W. I suggest 15W, 25W, 40W, 60W along with several 100W globes. You do not have to have them all plugged in at the same time. During early testing you might progressively switch in low values, later during testing they might all be 100 watt globes. 400W will only be needed at the end during reliability testing.

For 140Vdc output it would suggest …
watt, Rcold, Imax
15W, 240R, 0.6mA
25W, 144R, 1A
40W, 90R, 1.5A
60W, 60R, 2.3A
100W, 36R, 3.9A
400W, 9R, 15.5A. Avoid the 400W filament initially. Go for lower values, then several 100W globes.

The other thing you can do is put a light globe in series with the AC input during early testing. It should flash on momentarily while the storage capacitors initially charge, then go dull once running. If it stays on, then the load is too great or you have a short in the AC switching side of your inverter. The series globe should minimise the damage from catastrophic failures that take out multiple components. For example, if your half bridge fails, or one mosfet goes short, then the current through the remaining mosfet will be limited to below the series filament globe specification. You will clearly see if you have a problem.

I wonder does the soft start capacitor on the sg3525 acts only as a time delay depending on the capacitance used or does bigger capacitance actually startup the mosfet driving IC's slower? like more gradually.

See this application note, page 5/13;
http://freedatasheets.com/downloads... Supply And Power Management SG3525 AN250.pdf

On page 6 is the equation for “turn on time”, t = Css * 2V / 50uA. With a bigger Css It will take longer to start.

 

[Salvador]

now here's what's happening as of lately between me playing football and repairing my car.
i changed the IRFP460 to the newly received FCH104n60, now take into consideration that just before the IRFP's were driven by the same board which has the SG3525 and IR2110 on it , a 115w 230 bulb glowed nicely and even though i didnt try higher powers with the IRFP'S yet i soldered in the FCH ones.
now the same 115w bulb attached that made no problems to the IRFP and guess what happens when i switch on the power with the fairchild devices , nothing , not even a crack or a bang , but the lights went out immediately, so as I went to replace my fuses, came back and measured and yep there it was , both devices blown open completely , not physically but electrically , shorts on all three legs G D and S.

everything else was the same as before i changed nothing except the two half bridge mosfets driving the transformer.Mystery.
All in all I have to say I have never been a fan of semiconductors in places were high power both volts and amps have to be dealt with as any solid copper wire and a piece of iron will take any abuse and tolerate almoust anything but a semiconductor will not even close.tahst what i hate , yes I know that almoust every household item is now powered through a SMPS and their reliable and so on but well ehh I just have no luck with them , maybe because for them to work properly they need a heavy and advacned circuit were one parts watches over the other and so on and so forth and everything is designed and built to the maximum precision.

the other option is that the devices I have been given are fake and hence can't stand up to the requirments , since the IRFP's worked in the same situation with being lower rating devices.

maybe this will sound a little childish but since it's all just a big experiment i guess i will stick with what works and buy myself some more IRFP'S and just see how much power such a psu can give to my amplifier board , both with the satisfaction of my ears and also a small oscilloscope amplifier output load test. if they will do no good well FFS , i'll just use a toroid , atleast that thing aint going to break down in a hundred years.

 

[Baluncore]

, both devices blown open completely , not physically but electrically , shorts on all three legs G D and S.

That is a contradiction. Open or short ?
Did you have a 100W filament globe in series with the AC when you turned it on ?

I have never been a fan of semiconductors in places were high power both volts and amps have to be dealt with

High current is OK, high voltage is OK. But never both at the same time.
That may be your problem.
There is a problem somewhere in your circuit. I have insufficient information to guess where.

the other option is that the devices I have been given are fake

That is most unlikely.

 

[Salvador]

ups sorry yes , i was writing late at night got mixed up , both devices failed at the short circuit position showing less than 1 ohm on all legs no mater how you measure.
my bad i didint put the lightbulb in series with the incomming mains , maybe because i wasnt expecting something to happen since the load was small and it worked just fine with the previous IRFP's as before I changed them to the FCH I tested and it worked fine , after minutes of glowing the heatsink wasnt even warm it was cold and everything was fine.

I guess I'll just solder back my two leftover IRFP's and see it it works , after changing the IR2110 and probably the SG3525 which are blown everytime they mosfets get blown.
if it works then i will go to my friend , an electrician he has some lab instruments at work and just probe the waveforms coming out of the IR2110 into the mosfet gates and the transformer primary waveform and see maybe there is too small deadtime , maybe something else.
anyways it kinda weird.

 

[Salvador]

now I know it's hard to tell anything if one has to answer about a circuit thousands of miles away but , everything in my device is like in the schematic except the zener that feds the octocoupler is higher voltage and secondly the shunt across the primary winding is a ceramic 660pF with a 39 ohm resistor in series , instead of the 47 ohm and 470 pF ones in the schematic.
my question is is the shunt needed at all and if so is a series LC the best one to use in this case? maybe some high voltage spike developed in that shunt or something like that?maybe i could just use the 39ohm resistor and the 660pF caps in parallel?

even though i kinda doubt this was the cause but who knows.

 

[meBigGuy]

I think you need to test with a more constant resistive load, rather than a light bulb. The surge at startup coupled with a subtle issue could cause destruction.
Slower switching due to increased gate capacitance, for example.

Could even be the difference between typical specs and real world worst case values.

I'm not sure what the minimum load would be. Too light a load can cause failure also.

 

[Salvador]

thanks for coming into the discussion, yes i also kinda though of that maybe the fch104n60 devices have a higher gate capacitance since their also faster and higher rated devices than the IRFP's caused this.but then again was it because the IR2110 power supply sagged or the IR2110 can't or isn't menat to drive these devices (doubt this one)
i have to say I haven't seen in many smps a circuit that could or would sense frequency stability.

but since the lightbulb attached never came on yet the devices failed I have a feeling that maybe the IR2110 wasnt able to drive the mosfets , maybe the first was was left half open and once the second one opened it created a path and as all of the rectified mains current and voltage started to flow through it created a sudden and fast heat up and destruction of both devices.I sort of get that feeling because at the secondary side there was nothing at turn on not even a flash , so likely the transformer didint even see a flux set up.

I will check I actually have an old electrical heater made out of the same alloy wire used in high wattage big size wire wound resistors. I will measure the resisatnce and see if it could be used as a load.

speaking of too light of a load , i mean an amplifier with no input signal is basically almoust an open circuit except for some bias current especially in class A or AB. but mine is AB so would it even be safe to power an amplifier with this?
but then again what do commercial smps use for this as the basic principle is the same there is a transformer whose core if left with no load can act like an boost inductor causing spikes in voltage and so on.

 

[Baluncore]

my question is is the shunt needed at all and if so is a series LC the best one to use in this case? maybe some high voltage spike developed in that shunt or something like that?maybe i could just use the 39ohm resistor and the 660pF caps in parallel?

The series RC snubber looks like 47R for the high frequency transitions but appears to be open circuit when not switching. The 47R dampens the switching transitions, the 470pF reduces power consumption by the 47R at all other times.

If you used the R & C in parallel, the combination will look like a short circuit during transitions and like a heavy load at other times. The C will cause a current spike that will destroy the mosfet. Only the 1uF will limit the 47R resistor current, the 47R should be rated at maybe 2 watt * (1uF / 470pF) = 4.25 kW. Keep to the series circuit.

The published circuit you have is good. Do not make changes without good technical reasoning.
The problem you have is with implementation of that circuit. Maybe post a photo of your physical circuit layout.

If you want your components to survive testing.
1. Get a 100W filament globe in series with the AC supply.
2. Reduce the 4x 680uF to something like 2x 100uF during early low power testing. That will reduce the current spike on control failure.
3. Use a 40W filament globe as the load during early testing.

 

[Salvador]

I just checked the datasheets and if I am reading them correctly it turns out that the more powerful FCH104N60 actually has a lower gate charge to fully open the device.
both Qgs and Qgd are lower than the IRFP 460 , so I think it would be unlikely that the IR2110 which was driving the IRFP no problem could have had a hard time driving the FCH ones? I am correct in assuming this ?

 

[Baluncore]

Devices with lower gate charge are easier to drive and so can switch faster.
FCH104N60 was selected because it was better in all respects than original, older design.

 

[Salvador]

well that's what makes me wonder , the gate charge being lower and other aspects being better and it results in a short circuit.
well i forgot to mention that instead of 4.7 ohm in the gate drive path I have 10 ohm ones, i doubt though that it has any major effect on what's going on but what you think?

 

[Baluncore]

well that's what makes me wonder , the gate charge being lower and other aspects being better and it results in a short circuit.

Drain current and voltage specifications need to be greater than or equal to the original design. Higher current mosfets take a greater area of silicon and so they tend to have greater capacitance with a greater charge to move during transition. Minimising gate charge is required to get a faster transition. The design of new mosfets is to lower the gate charge while increasing the maximum drain current.

well i forgot to mention that instead of 4.7 ohm in the gate drive path I have 10 ohm ones, i doubt though that it has any major effect on what's going on but what you think?

The 4R7 is there to stop a parasitic gate oscillation. By using 10R you have still killed that parasitic but you have also slowed down the transition which could cause overlap of the conduction. So it could be a problem.

Good high frequency layout is essential and critical to reliable SMPS design. For example, you can expect parasitic RF oscillation of the gate drive circuit if the distance between the IR2110 and the mosfet gate is greater than about 5 cm. How long and how thick are your floating common reference traces or do you use a local ground plane?

You have still not provided a photo of the layout of your implementation. We can still only guess at the cause of your problem.

 

[Salvador]

I will take a photo as soon as I will get a camera to do so, maybe in the weekend.since i threw this thing together much from what I had in hand and a little bit of "macgyver" thinking , the whole device consists of three parts , firt part is the mainboard which has the mains rectifier , capacitors and a small transformer for the IC's power supply, and which also has the other parts attached to it with physical connections.
then i have a separate small board which has the oscillator and control IC (SG3525) and the driver IC (IR2110) with all the periphery those IC's need on that board.then i have the heatsink with the mosfets on it next to that board and the connections to that board are made with short wires., everything else is the transformer and then there is the secondary side with the rectifier and so on , not the prettiest looking board for sure but everythings there.

now here's the question , before i touch the 10ohm gate resistor , does it really matter here ? because today as I looked at the datasheets and I think you confirmed that , that the FCH has a lower total charge (both Qgs and Qgd) so if the IRFP's have a higher charge then it should be the other way around , the system should fail driving the IRFP's not the FCH yet it goes vice versa.

now this evening i did some more tests using just one of the two onboard half bridge smps (basically just two of these circuits on one piece of a board,)
i had put the IRFP 460 back just to see what happens and guess what , I now put two bulbs in parallel , one was 120w the other was 57w
measured the cold resistance of these parallel bulbs and it was about 22 ohms.when i turned on the smps it worked with no problems whatsoever.nothing got warm no voltage fluctuations, no nothing.
How can this be , there must be some hidden obstacle here why the FCh devices failed wthout even giving me a glow on the secondary sides atached bulb, the previous time the IRFP's failed due to the much too big 500w halogen bulb , atleast the bulb got a bit bright before the IRFP'S reduced themselves to the ash they became.
(If I keep on like this I'll soon have a full container of copper to go sell to the recyclers , atleast i'll get some money back :D)

Anyways since the IRFP'S acost 1/3 of the FCH devices I will test them some more to see what they can and maybe i'll just stick to them for now , I hope on lighting up the 230w halogen bulb which has a cold resistance of 15 ohms , it they can do that I'm satysfied.
now the other options for testing I have (please don't laugh) is a sandwitch toaster which measures about 80 ohms of resistance , and then i have an old wire heater (basically a high resistance wire wound around a ceramic tube) the resistance cannot be measured because the wire is old and the contacts being metal are corroded from the effects of heat. with a multimeter it shows about 6Kohms which i assume can't be true since then its power would be very small but its somewhere between 600 and 900w.
Ok one more question suppose my IRFP's work happily cooking me bread in a toaster or heating my room in a dummy heater load, but what would they do with driving an actual amplifier, now i don't know what te resistance is as seen by the psu, when one listens to music on low levels since the transistors are always only 1/3 open or so but what would happen if i used a powerful speaker at high volume, assuming the speaker is either 4 or max 8 ohms , i mean what would the PSU see as its resistance ?
because I know what it does when it's sees a cold 10 ohm filament for a split second , it disintegrates.

 

[Baluncore]

For SMPS design and construction help, download this pdf; http://www.ti.com/lit/ml/slup224/slup224.pdf
Take a look at topic 4=Layout, topic 1=safety, and 3=AC line.

 

[Salvador]

I am reading this pdf now, while at it I have a question that I couldn't find doing google searches.
when the psu is connected to an amplifier load, aka drives an amplifier, surely the load of an amplifier is not resistive only but it's an impedance load , so in this case waht would be the average load seen by the power supply driivng the load through an amplifier?
surely i understand it would change constantly since music is a signal with varying waveform, but just assuming some averages here , given that the amplifier board can take about 500+ watts at 4 ohms , assuming its driving a decent speaker load at those 4 ohms.
I guess its kinda complicated to arrive at the answer , yet it will determine if my psu will go up in smoke again if i turn some club hit loudly on a big sized woofer... just for example.

P.S. the 230w 230 volt halogen bulb test also comes out as successful , runs with no problems ,as much as i could measure the filament when cold it was about 16 ohms. which gives me the starting current at 140 volts DC of about 9.3 amps.

 

[Baluncore]

surely the load of an amplifier is not resistive only but it's an impedance load , so in this case waht would be the average load seen by the power supply driivng the load through an amplifier?

A SMPS sees it's output capacitance. The SMPS regulates the voltage on that capacitance, as best it can.
The SMPS must provide the required charge to maintain capacitor voltage.

The amplifier sees that same capacitance, with a fixed voltage, and so can draw the current it needs.
A regulated capacitor voltage Vc, with a variable amplifier current requirement i, represents a variable resistance load R = Vc / i.

The dynamic impedance of the amplifier is isolated from the SMPS by the intermediate storage capacitance.

 

[Salvador]

the actual running of the smps is poor, it doesn't drop volts when running with the lightbulb but that is only because its operational white hot filament resistance is high and it drawn only about 0.8 amps at 140 volts DC.
I was testing today using resistive dummy loads and an oscillator , the gate drive waveform appears to be good , nice symmetrical square wave , .firstly i tested loads from 40 ohms to 20 ohms and all of them made the output sag, at 20 ohms the output went from 140 vDC to about 110vDC, and the higher in resistance i went the lower it got yet still.
now in the schematic i measured the frequency putting the probes first for the upper transistor at leg7 and leg 5 on the IR2110, then for the lower one leg1 and ground.
in both cases i got about 1 square wave for about 20uS.

now if my maths is right there are a million uS in one second so dividing those million Us with the 20uS i got one period at the gate drive turns out to be 50khz , so the frequency is about right, also the duty cycle seems to be correct,
is it possible that having too many turns on the primary could result in less power considering this topology? i just saw an old paper after which i made the transformer and it has a little over 30 turns for the primary if I am correct.

 

[Baluncore]

at 20 ohms the output went from 140 vDC to about 110vDC, and the higher in resistance i went the lower it got yet still.

Which indicates that the system is not regulating duty cycle correctly, or that there is a limitation in the energy flow path.
Maybe your transformer is limiting energy flow? Reduce the number of turns to detect if saturation of the magnetic core is a problem.

 

[Baluncore]

Attached here is a useful reference for use when designing SMPS transformers.

 

[Baluncore]

Here are more. Everyone has a different approach. Get all these files, then browse them until they start to make sense.

slup123 “Magnetics Design for Switching Power Supplies”; http://www.ti.com/lit/ml/slup123/slup123.pdf

slup171 "Magnetic Field Evaluation in Transformers and Inductors". http://www.ti.com/lit/ml/slup171/slup171.pdf

Attached is; slup205 "Transformer and Inductor Design for Optimum Circuit Performance"

 

[Salvador]

OK, now I have a lot of material will take me a few days to get through it all with a cool head.
meanwhile i can say that he first smps (two of them being on my board) is running with a "3c85" rated ferrite core, I am not sure i'll try to look up maybe judging by this ferrite index we can approximately know how many turns would be required for a given frequency since we know the ferrite brand.

also what makes me wondr is that I was using a 82v zener for the feedback, since i though that i could have more voltage on he rails since the amplifier can take it.
yet still I am only doing +-70 volts DC output at max. so maybe i should just get a 66 volt zener and see what happens , the octocoupler should work then and maybe youre right , maybe the pulse width isn't regulating.it probably isnt.
i will have more oscilloscope tests next week then i will also try to measure the duty cycle with different loads.
it has to be the transformer that limits the current , the output diodes used are MUR1560, the driving circit is working fine.
maybe the fault lies in the transformer design , still a mystery why the FCH devices failed

 

[Baluncore]

yet still I am only doing +-70 volts DC output at max.

If the output is not regulated then do not mess with the feedback circuit.
You should instead lower the test load and identify the reason why the supply has such poor power output that it will not regulate.

maybe judging by this ferrite index we can approximately know how many turns would be required for a given frequency

That core is for a wound transformer, NOT an inductor. Frequency is determined by the components on the PWM clock.
The important thing now with the transformer is the ampere*turns which could be causing current limiting by saturation of the core.
The inductance of the transformer determines magnetisation current.
You need as many turns as possible without core saturation at maximum output current, so as to keep the magnetising current as low as possible.

Keep reading the references until they start to make sense and you can put numbers in the equations.

 

[mheslep]

You must be careful testing that circuit. I use a 1:1 mains isolation transformer and Earth the common reference when working inside live switching power supplies.

Yes, that's step two or three for me. Before getting anywhere close to the mains voltage, I'll run the primary off of a low voltage AC bench supply and the control circuitry off a bench DC supply, with the feedback loop jiggered accordingly. Later with with the mains isolation transformer in place, first use a well below design rating fuse, and so on. Little by little.

 

[Salvador]

ok here's what i think i should do , take a 0.1 resistor and place it in series with primary winding then attach scope probes across the resistor and see the waveform under the condition when the secondary voltage sags abut 10 to 20 volts under load. this should show me whether the core is saturating or not, even though i read that in half bridge the core should't saturate under load if it hadnt saturated at idle already, kinda makes sense because as you add load to the secondary it becomes harder for the primary to keep the same magnetizing current as it did with no secondary load isn't this true?

also i think i should insert the 66v zener instead of the 82 volt one and then drive the smps under the same load and monitor the signal on the gates of the mosfets with the scope to see if the waveform chages under load (whether PWN works or not)

also if my calculations are correct the supply did 110 volts instead of 140 idle when it had a 20 ohm resistor as its secondary load , a simple calculation shows that V/R = 110/20 =5.5 amps , now take 110 volts times 5.5 amps turns out to be just a little over 600w of power delivered to those resistors at that time and load, which isn't exactly small and maybe isn't my goal either , so right now it seems that either core saturation or non working PWN could be the cause for the drop , as if i would maintain 140 volts with a 20 ohm load , the power would be 980 watts , and quite frankly that would be way too good for two IRFP 460 driving a homemade half bridge.
so maybe its not that bad after all.
ok I will continue on reading the materials provided and other information and also measure and test my supply to see where the falt may lie, also the destruction of the FCH devices is still an open case.
any information from you folks is much appreciated and also I want to say thank you for the help already given on this topic.

 

[Salvador]

also , i think I will increase or at first for experimental purposes use a reostat, in place of the resistor between sg3525 pins 7 and 5, now there is a value of 33 ohms as shown in the original schematic , but with this low value the deadtime mayb be too short and given that i have more turns than originally shwn on primary may have been the cause for the FCH untimely death.
i also saw on scope when i probed the gate signals that there is close to no deadtime.my plan is to increase deadtime with this resistor and then after tests increase frequency with the decreasing of the other resistor for which i already have a reostat inserted and then will measure the voltage response to applied load.

 

[mheslep]

What kind of scope probe do you intend to connect to the primary?

 

[mheslep]

Jim Williams of Linear Tech has the definitive smps design guide, imo.
http://www.linear.com/docs/4120
See appendix C in particular for a safe method of testing out your magnetics and switches for saturation prior to assembling the entire design.

 

[Salvador]

I see that in the west you are extremely concerned about safety , I mean it's not like I or anyone else I know measures the potentials with his fingers or probes a live switching circuit while washing his body in a bathtub full of water and a nice long earthing cable attached to it.
and while on this topic , quite frankly I have seen safety regulations being abandoned much more by professionals than amateurs like me , because when i get to a place where once in a while I am not certain what might be waiting at the other side i use a safety tool instead of my hand to check things or sometimes when i have worked with live AC at the wall installation just those rubber gloves that insulate from electricity also stepping on a rubber pad combined with the gloves does the trick quite nicely.as for the question you asked , I will be using as I said alow value resistor in series with the primary so that i can attach an oscilloscope and see the waveform , i mena i could do it otherways but then i have to use some resistors in other way because its an old but simple soviet made scope and the max you can see in it is about 50 volts +-.
the man with whom I am working with has repaired tons of tv's back in the day using that and other scopes so we will just probe the primary to see what going on with the waveform.
also the deadtime , i have to make it longer at first to put less stress on the mosfets and to save them from a potential current shooting through.
also will see whether the length of each pulse has to do with driving the core into saturation which maybe limiting the power output.