Lowering SMPS dissipation

[dzseki]

Here is an excerpt of an SMPS circuit. The design comes from the early 90’s. Q20 being the switching element is mounted on a rather small heatsink because of size constraits (possible design flaw, possibly could have done better) and the thing is running hot obviously. Utilizing a modern MOSFET (with much smaller RDSon) would be a good way to reduce the dissipation there.
I am not the first one tempted by this idea in this particular circuit however, other tinkerers have changed the Q20 MOSFET to IRF740 but the reported behaviour was the FET went short rather soon.
I haven’t tried this for myself, and I find it somewhat surprising. Gate capacitance is important for sure but the MC34151 should have no problem with driving that, what other parameters should be observed then?

[spec]

Hi dzseki,

Just thinking out loud really.

You mention that the heatsink is not really adequate so considering that area:

Are you certain that you can't beef up the heatsink one way or another?
Look for an NMOSFET with a lower thermal resistance junction to case, as you mention.
Use the best new thermal grease.
Investigate thermal insulating washers with a low thermal resistance- I think aluminum oxide is the lowest, but ceramic is also pretty good.
In the same vein, if the transistors are T0220, consider a case clamp to lower thermal resistance to the heatsink or change to TO3.
You could consider making the heatsink live and thus eliminating the insulating washer completely. This should reduce the thermal resistance to the heatsink a fair amount, but don't know how practical this would be. You would have to be cautious about the extra capacitance and inductance this would introduce to the drain circuit too.
Fit a fan- this is not as radical as it sounds- you can use the small CPU type fans which are compact, but effective.

The MC34151/33151 driver chip has a fairly high drive current capability at 1A, but that is not an awful amount by modern standards, where some drivers can sink and source 4A to 7A. Modern MOSFETs with lower RDss, tend to have higher parasitic capacitances in general.
There is a spare section of the driver, so maybe you could parallel them up to get more drive. It would be best to fit a better driver though if you were seeking more drive current. But, the thing is that driving MOSFET gates is a complex task with quite a few variables to consider.
Fit a fan- this is not as radical as it sounds- you can use the small CPU type fans which are small, quiet, and compact, but effective.

And a real crazy idea- fit liquid cooling as used on some high end PCs. I have looked at this approach for a few critical applications and it is possible to design quite a simple, low cost system with a bit of ingenuity.  Last time I looked, off-the-shelf liquid cooling kits were costly.

You could consider an altogether more beefy NMOSFET- IXUS make some real beasts.

Check and possibly upgrade the critical capacitors may be worth considering too.

And that is about it.
 

[T3sl4co1l]

Yeah well, IRF740 isn't much improved from what they've got there...

Actually it's pretty damn close anyway.  If the original design is that borderline, I guess that would do it.

Try something newer like an, uh, FDPF12N50T or IPAN50R500CEXKSA1 or STP12N50M2 or something.

Could try replacing D21, D22 with schottky e.g. C3D03060F or something.  It's probably not running in hard switching so this shouldn't be a big deal.

The snub circuit (D23, R94, C56, etc.) could be changed too, depending on what all is happening.  In particular, a dV/dt snubber (R94 --> 0.0, place a ~1k in parallel with D23, C56 --> 1nF C0G maybe) can prove helpful for saving switching losses.

Mind, all these changes are contingent on the EMC remaining acceptable, both in terms of internal signals (don't want lines/dots/crawl in the video?) as well as external (emissions).  The latter is a bit harder to evaluate, and I might just as well suggest finding a useful supply inside (like the 6V output there, or a 12 or 24V supply elsewhere, maybe?) and running a fan off that.  Can also try shopping around for better heatsinks in the same space / footprint, or adding on more material to improve what's there, utilizing whatever nearby free space you can find.

Tim

[T3sl4co1l]

Are you certain that you can't beef up the heatsink one way or another?
Look for an NMOSFET with a lower thermal resistance junction to case, as you mention.
Use the best new thermal grease.
Investigate thermal insulating washers with a low thermal resistance- I think aluminum oxide is the lowest, but ceramic is also pretty good.

It sounds like it's a small on-board heatsink, in which case, none of these will be much use unfortunately -- RthHA (heatsink-ambient) is dominant by a huge amount.  Some kind of grease is still needed (a dry joint can be RthCH ~ 10 C/W easily, but greased is typically ~0.3), but beyond that, you're just beating yourself up.

In the same vein, if the transistors are T0220, consider a case clamp to lower thermal resistance to the heatsink or change to TO3.

I would recommend against moving the transistor somewhere.  Stray lead length will cause RF oscillations and slower switching.

Tacking stuff onto the heatsink to make it bigger, or connect it (thermally) to the enclosure, is a good idea though!

You could consider making the heatsink live and thus eliminating the insulating washer completely.

...Speaking of -- it may already be, so watch out if you add anything to it, don't accidentally short things out!

The MC34151/33151 driver chip has a fairly high drive current capability at 1A, but that is not an awful amount by modern standards, where some drivers can sink and source 4A to 7A. Modern MOSFETs with lower RDss, tend to have higher parasitic capacitances in general.

There is a spare section of the driver, so maybe you could parallel them up to get more drive.

Wouldn't worry about it.  Paralleling is not a bad thought.  Wouldn't bother with anything bigger.

My power supplies using UC3842 (which has a very similar gate driver section) are just fine with transistors about this size.  Difference of course being, I have enough heatsinking for the transistors to survive.

Fit a fan- this is not as radical as it sounds- you can use the small CPU type fans which are small and compact, but effective.

Yup, even if it's just improving airflow within an enclosure, it's still something.  If it draws in more airflow through vent holes and such, even better!

Downside: it needs cleaning every once in a while, as the forced air deposits dust everywhere.  (Surprisingly, natural convection can stay very clean -- I have a CRT monitor that's been used for years and years, and it's still very clean inside.  Anything with a fan?  Gunked up in a few years.)

And a real crazy idea- fit liquid cooling as used on some high end PCs. I have looked at this approach for a few critical applications and it is possible to design quite a simple, low cost system with a bit of ingenuity.  Last time I looked, off-the-shelf liquid cooling kits were costly.

Hah!  I wouldn't worry about it for anything less than a couple kilowatts -- but if you had it handy and want to put the time and materials into it... why the heck not, I guess?

Might even be better than that, if it's a projector as it says, it may already have water cooling to deal with the lamps / CRTs / optics.  Tap into a nearby hose and there you go -- just mind the liquid conductivity, don't want that nasty high voltage and noise getting into everything else.

Tim

[xavier60]

Also there are some odd things about the schematic. The values of R90 and R94 look wrong. The position of the snubber components under the transformer don't make much sense. Maybe C54 goes to ground?
EDIT: R90 could not possibly really go to Drain!

[xavier60]

Failure of C50, R216, R217, R94 would stress Q20. Has the problem always existed from new?

[dzseki]

Thank you guys for the responses.
A little more info,
The board slides in a metal cage that has forced air cooling. Q20 is located very near to the air inlet, although not the shape nor the orientation of the small heatsink (yes it is a small PCB heatsink) is optimal for this cooling, but it gets considerable draft so it kinda works anyway.
The design is not completely bad, as it is working well in most cases, the problem arises when a user replaces the fans for quieter operation, we all know this is a red flag, but a desire for a quieter operation is also understandable, and this thread is aiming to find out if anything happened in the world in the past ~25 years to improve the efficiency the circuit.
This SMPS is located on the horizontal deflection module, and unlike to CRT monitors and TV this circuit was made rather beefy (it permits horizontal scanrates in excess of 150kHz with retrace times well below 2us on 3 deflection coils simultaneously), so there is a nice big heatsink already on the board where the deflection MOSFETs (and other transistors)are mounted (and the forced air is mainly aimed on this heatsink). For whatever reason this Q20 does not mounted on the big heatsink, instead it has its own small in close proximity to the big one (say 2" apart). The heatsinks are isolated from the drain and are grounded.

EMI is a concern here already, the original design is what I posted here. Years later there was an upgrade to the circuit and more snubbing was added, like shown. At the same time the heatsink for Q20 was replaced with a somewhat larger one, but I believe the sole reason for this was the even increased dissipation of the FET because of the added gate drive snubbing.

I also attached a picture of the physical board, Q20 is mounted on the black heatsink, and the air comes from the direction where we see the board, not very effective.

I was looking at SiHF15N60E MOSFET as a substitute, sure the snubbing should be adjusted accordingly.

[xavier60]

The extra Gate components are to reduce the MOSFET's switching speed, to reduce EMI.
They have slowed the Off speed by about half and the On speed by about 20 times. The On speed with just the 10Ω resistor would be way to fast anyway. This will add to the switching loss but it is complicated to say by how much. For example if it is operating in Discontinuous mode, slowing the On speed makes less difference to switching loss compared to Continuous Conduction mode.
Conduction loss is a function of the current waveform and resistance.
I think that SiHP15N60E would make worth while improvement to conduction loss. It will switch faster due to its lower Charge specs and possibly worsen EMI.
I chose to reduced the switching speed of the power supply MOSFET in my HP 54645A DSO to reduce EMI.

[spec]

Q20 is located very near to the air inlet, although not the shape nor the orientation of the small heatsink (yes it is a small PCB heatsink) black heatsink, and the air comes from the direction where we see the board, not very effective.

I would think a lot could be done, quite simply, to improve the cooling of Q20, considering how minimal that little heatsink is. What is the Q20 failure mode? Is it a short circuit D/S, indicating overheating.

I was looking at SiHF15N60E MOSFET as a substitute, sure the snubbing should be adjusted accordingly.

Nice find- I have added that to my list of HV NMOSFETs.

[spec]

Are you certain that you can't beef up the heatsink one way or another?
Look for an NMOSFET with a lower thermal resistance junction to case, as you mention.
Use the best new thermal grease.
Investigate thermal insulating washers with a low thermal resistance- I think aluminum oxide is the lowest, but ceramic is also pretty good.

It sounds like it's a small on-board heatsink, in which case, none of these will be much use unfortunately -- RthHA (heatsink-ambient) is dominant by a huge amount.  Some kind of grease is still needed (a dry joint can be RthCH ~ 10 C/W easily, but greased is typically ~0.3), but beyond that, you're just beating yourself up.

In the same vein, if the transistors are T0220, consider a case clamp to lower thermal resistance to the heatsink or change to TO3.

I would recommend against moving the transistor somewhere.  Stray lead length will cause RF oscillations and slower switching.

Tacking stuff onto the heatsink to make it bigger, or connect it (thermally) to the enclosure, is a good idea though!

You could consider making the heatsink live and thus eliminating the insulating washer completely.

...Speaking of -- it may already be, so watch out if you add anything to it, don't accidentally short things out!

The MC34151/33151 driver chip has a fairly high drive current capability at 1A, but that is not an awful amount by modern standards, where some drivers can sink and source 4A to 7A. Modern MOSFETs with lower RDss, tend to have higher parasitic capacitances in general.

There is a spare section of the driver, so maybe you could parallel them up to get more drive.

Wouldn't worry about it.  Paralleling is not a bad thought.  Wouldn't bother with anything bigger.

My power supplies using UC3842 (which has a very similar gate driver section) are just fine with transistors about this size.  Difference of course being, I have enough heatsinking for the transistors to survive.

Fit a fan- this is not as radical as it sounds- you can use the small CPU type fans which are small and compact, but effective.

Yup, even if it's just improving airflow within an enclosure, it's still something.  If it draws in more airflow through vent holes and such, even better!

Downside: it needs cleaning every once in a while, as the forced air deposits dust everywhere.  (Surprisingly, natural convection can stay very clean -- I have a CRT monitor that's been used for years and years, and it's still very clean inside.  Anything with a fan?  Gunked up in a few years.)

And a real crazy idea- fit liquid cooling as used on some high end PCs. I have looked at this approach for a few critical applications and it is possible to design quite a simple, low cost system with a bit of ingenuity.  Last time I looked, off-the-shelf liquid cooling kits were costly.

Hah!  I wouldn't worry about it for anything less than a couple kilowatts -- but if you had it handy and want to put the time and materials into it... why the heck not, I guess?

Might even be better than that, if it's a projector as it says, it may already have water cooling to deal with the lamps / CRTs / optics.  Tap into a nearby hose and there you go -- just mind the liquid conductivity, don't want that nasty high voltage and noise getting into everything else.

Tim

Boy- You have been busy Tim.

[xavier60]

If Q20 is moved to the large heatsink, the capacitive coupling between the tab and the heatsink can be a great cause of EMI.
Those thick aluminum oxide washers  would help a lot. I would use a SIHP15N60E with another 10Ω at the Gate pin.
https://uk.farnell.com/aavid-thermalloy/4171g/insulator-alum-oxide-ceramic-to/dp/2787815
When Q20 fails, how much damage is caused?
With power supplies where the shunt resistors blow open, it makes a real mess.

[spec]

If Q20 is moved to the large heatsink, the capacitive coupling between the tab and the heatsink can be a great cause of EMI.
Those thick aluminum oxide washers  would help a lot. I would use a SIHP15N60E with another 10Ω at the Gate pin.
https://uk.farnell.com/aavid-thermalloy/4171g/insulator-alum-oxide-ceramic-to/dp/2787815
When Q20 fails, how much damage is caused?
With power supplies where the shunt resistors blow open, it makes a real mess.

I did not say that Q20 should be moved to a larger heatsink if that is what you are implying. Q20 can be kept pretty much where it is but the heat sinking can still be radically improved, I would suggest. But even so, I doubt that the situation is that critical.

On the subject of inductance etc, it would be useful to establish what trace lengths, and thus inductance etc exist on the board as it is now.

[xavier60]

The op mentioned "there is a nice big heatsink already on the board" earlier. It has some risk.

[spec]

The op mentioned "there is a nice big heatsink already on the board" earlier. It has some risk.

I see!
One approach to make use of the large heat sink, without significantly moving Q20, is to fabricate an aluminum bridge between Q20's existing heatsink, which to put it bluntly is a joke, and the large heatsink. But rather than that, I would think that it would be best to get rid of Q20's existing heat sink and either buy or fabricate a  completely new, but far lower thermal resistance heatsink. A bigger standard heat sink may fit- I haven't checked- or a standard heatsink modified may do. The revised  heatsink may or may not bridge to the bigger heatsink. It would almost certainly need to be supported by the large heatsink.

But in any case, I see this as a simple, low-cost approach to improve Q20 cooling, with minimal technical risk. Of course, a better NMOSFET can be explored as a parallel activity, but, as you illustrate, it is a complex task.

I have done a bit of work on driving MOSFETs and EMC certification, but not to the extent that you obviously have, and I find your detailed explanations, not just on this thread, quite interesting and helpful. It is a difficult area which crops up often and does not seem to be well understood. The opposing requirements of MOSFET fast turn on/off and EMC further complicate the issue. Have you ever thought about doing a simple tutorial for EEV?

[xavier60]

I think that the  SiHP15N60E should be used also. If more EMI is noticed, the Gate resistor value can be increased.

[dzseki]

When Q20 gives up it conducts through D-S, obviously taking the current sense resistors as well, not much damage other than that.

The FET is located in the bottom of the cage where the board slides in, so you can't access with a scope probe to check things while working, adjusting the gate drive has to be done "blindly", but I may try anyway. For that I should record at least the heatsink temperatures before/after to have any meaningfull output from the experiment, other than possible EMI problems.

Regarding the SiHP FET it has almost double Gate capacitance than the IRF730 so wouldn't that inherently slow down the switching speed further with the given drive resistors (let's consider the one with the 220Ohm in it), which would also ease the EMI, but at the expense of possible increased dissipation?

I also consider the possibility of using custom heatsink, like ones with nice fins you can find on DDR chips on video cards for example, with fins aligned to the air flow directon, and only a "translator" element has to be made between the transistor's Drain and the plane of the heatsink, which would be perpendicular.

[xavier60]

When Q20 gives up it conducts through D-S, obviously taking the current sense resistors as well, not much damage other than that.

The FET is located in the bottom of the cage where the board slides in, so you can't access with a scope probe to check things while working, adjusting the gate drive has to be done "blindly", but I may try anyway. For that I should record at least the heatsink temperatures before/after to have any meaningfull output from the experiment, other than possible EMI problems.

Regarding the SiHP FET it has almost double Gate capacitance than the IRF730 so wouldn't that inherently slow down the switching speed further with the given drive resistors (let's consider the one with the 220Ohm in it), which would also ease the EMI, but at the expense of possible increased dissipation?

I also consider the possibility of using custom heatsink, like ones with nice fins you can find on DDR chips on video cards for example, with fins aligned to the air flow directon, and only a "translator" element has to be made between the transistor's Drain and the plane of the heatsink, which would be perpendicular.

I hoped that R97 would blow first. The messy aspect is that damaged parts can get missed, for example, I see gate resistors go high resistance. The power supply will work for a while then fail again.
I was comparing to the IRF740. Gate to Drain charge has the most effect on switching speed. It also adds directly to the losses as the MOSFET has to discharge itself every time it turns on.
I'm not familiar with the DDR heatsinks.
The drive mod with the 220Ω might not increases losses much at all depending on the operating mode of the supply, which is unknown and it's likely very reliable with the original fan speed.

[NiHaoMike]

What about use the lowest gate resistance that doesn't ring to minimize switching losses and then add filtering to solve EMi problems?

[xavier60]

Although there are many unknowns, looks like the control IC is configured to limit the MOSFET current to about 2A peak.
Assuming DCM and 50% duty cycle, worst case MOSFET RMS current is 0.6A. And because it's running from 85V rather than 340V that I'm used to, the switching losses would be lower unless this power supply is running at an unusually high frequency.
C50 should be checked, just in case.
Does R90 really go to where it's shown?

[xavier60]

Rt and Ct to me read 7.5K and 1nF on pins 18 and 16 of the TDA4718A.
When I try to apply these values to the VCO frequency chart on page 15 of the TDA4718A data sheet, the result is literally off the chart, very high.
Can someone confirm?

[dzseki]

What about use the lowest gate resistance that doesn't ring to minimize switching losses and then add filtering to solve EMi problems?

As it is this is a good idea, however when the board is in its place Q20 is literally unreachable with a scope probe, so this encumbers the effort.

Rt and Ct to me read 7.5K and 1nF on pins 18 and 16 of the TDA4718A.
When I try to apply these values to the VCO frequency chart on page 15 of the TDA4718A data sheet, the result is literally off the chart, very high.
Can someone confirm?

On my copy it is still on chart, and it is somewhat beyond 100kHz.

[xavier60]

On my copy it is still on chart, and it is somewhat beyond 100kHz.

Yes, I misread it. Still high, so switching losses are significant.

[dzseki]

So I made a custom heatsink. I've checked the ready made ones, but wasn't too impressed by any of them... So for this round the circuit remains unaltered only the heatsink was changed.
PS.: I haven't tried the board yet...