Tips for reducing EMI caused by PFC boost diode reverse recovery?

[TimNJ]

Hi all,

I'm working on ~200W, small form-factor AC/DC converter. I'm having some common-mode conducted EMI issues at approximately 15MHz or so. Looking at the voltage across the PFC boost diode, I noticed the reverse recovery time (trr) is approximately half the period of 15MHz, leading me to believe that it is the culprit. (Not 100% certain, though).



The board is very space constrained and I'm not sure if we can squeeze many more turns on the AC common-mode choke, but it is certainly worth a shot. Maybe a different core material?

Perhaps we could swap the PFC boost diode for a diode with a "softer" recovery characteristic? What specifications are most important for that? I was thinking reverse recovery charge (Qrr).

Here's the diode currently on the board: https://www.vishay.com/docs/93244/vs-8ewx06fn-m3.pdf

Does anyone have any tips/tricks? Could use all the help I can get!

Thanks a lot.

[Gyro]

Does anyone have any tips/tricks? Could use all the help I can get!

An RC snubber across the diode would be my first course, at least to confirm your suspicion that it is the culprit.

[David Hess]

A softer recovery diode will help if that is the problem.  Sometimes a snubber is added to the diode.  I have even seen an EMI bead added to one of the diode's leads.

Check out Linear Technology application note 118 for some ideas.  At the end it includes the designs for a simple h-field EMI sniffer probe and how to use it.

[Siwastaja]

A SiC diode would be the modern answer to this problem. Even though more expensive, reducing power losses, getting rid of additional snubber losses, and getting rid of the extra complexity of snubber design and measurement is often worth it.

If not, then look at both reverse recovery time and softness. If the diode recovers quickly enough, "softness" may not be needed. If, on the other hand, the "soft" diode is too slow, it's still worse than a "hard" fast diode. In most cases I have encountered, the "soft" diodes I tested were, at the end of the day, inferior because they were much slower (think about 60ns instead of 15ns). This means the shoot-through current level (which increases with time, limited by parasitic inductance) was much higher (think about 50A instead of 10A), meaning high losses. And coming out of this huge current peak is not good for EMI, even if it happens somewhat slower - and this softness means you are wasting even more energy during reverse recovery than what the trr value tells you. So, softness is definitely good but only if it doesn't mean compromising on the opposite side of the very same issue.

Sometimes a simple RC snubber is like magic, and not necessarily wasting that much energy. It's best to try out. Unlike the literature and appnotes teach you, I design the RC snubbers by choosing the C value based simply on my power budget available (limited by the resistor case size, or maximum efficiency drop I'm starting to get uncomfortable with), then just test.

[nctnico]

An ferrite bead works well too, is easier to design in and I suspect it has less losses than a snubber. Just make sure the ferrite bead is choosen well when it comes to the frequency response.

[T3sl4co1l]

An ferrite bead works well too, is easier to design in and I suspect it has less losses than a snubber. Just make sure the ferrite bead is choosen well when it comes to the frequency response.

+1 this, but with caveat:

You may very well be at high enough power level that you get a Metcal tip rather than an EMI bead.

That is to say, the ferrite bead's losses (to your circuit, it's basically a saturating L || R) are high enough that its temperature rises to Tc (Curie temperature), where L drops off, losses drop off, and it regulates its temperature.  Meanwhile, your EMI spectrum looks good for the first few seconds of operation, then the spikes rise back up to normal levels.  A hot ferrite bead is a useless ferrite bead.

Solution: either reduce loop inductance, increase commutation time (losses go up), or snub the loop inductance.  Which, since reduction is the first priority and you're unable to do that, you now have to increase it, by adding an inductor (not a ferrite bead -- inductors saturate at rated current, ferrite beads saturate at far less!), and damping or clamping that.

Also, have you done the dance with the Y caps?  Sometimes you want one pri-sec or pri-GND, other times you want as little as possible.  Depends on the exact windup of the transformer, and whatever else is farting out noise (heatsink capacitances, proximity, etc.).  I always put in all combinations, in the PCB design, then try permutations until I've found the minimal set and values.

Assuming, of course, a traditional PFC --> converter design.  If N/A, then YMMV.

I suppose if things are very compact, you could also have induction into the supposed-to-be-filtered side, which can really only be solved by reducing loop areas (use 2+ layer construction??) or making it bigger (getting the filtered side away from the noisy side).

Tim

[T3sl4co1l]

Also, you're right to suspect that a couple more turns on the CMC might not do very much.  That's up in the cutoff range of most CMCs -- that is, when there's a single bulk winding, the turn-turn capacitance dominates.

Try a CMC with fewer turns, and more sections (usually bank-wound like these,
https://www.digikey.com/product-detail/en/epcos-tdk/B82733V2701B001/495-7017-ND/4945303
https://www.digikey.com/product-detail/en/schaffner-emc-inc/EV28-1.5-02-20M/817-1046-ND/1928625
notice each winding is in two sections with a divider between), or higher permeability core (nanocrystalline, i.e.
https://www.mouser.com/ProductDetail/Vacuumschmelze/T60405-R6131-X037?qs=sGAEpiMZZMsVJzu5wKIZCQ7jTXqOmi27XOwRTjSerLI%3d
and such), or managing to squeeze in one more tiny choke that takes over at the high frequency range (unfortunately, these aren't common, and they usually aren't as compact as their values would imply ).

Tim

[TimNJ]

Does anyone have any tips/tricks? Could use all the help I can get!

An RC snubber across the diode would be my first course, at least to confirm your suspicion that it is the culprit.

Thanks! Got about 6dB reduction @ 15MHz with addition of a snubber. I didn't try much to fine tune it, and I think it is dissipating a lot of power. However, as you suggest, it proves that the diode is the culprit.

[TimNJ]

A softer recovery diode will help if that is the problem.  Sometimes a snubber is added to the diode.  I have even seen an EMI bead added to one of the diode's leads.

Check out Linear Technology application note 118 for some ideas.  At the end it includes the designs for a simple h-field EMI sniffer probe and how to use it.

Thanks for the link. I think I'll try to build this up at some point. Looks like it could really come in handy.

[TimNJ]

A SiC diode would be the modern answer to this problem. Even though more expensive, reducing power losses, getting rid of additional snubber losses, and getting rid of the extra complexity of snubber design and measurement is often worth it.

If not, then look at both reverse recovery time and softness. If the diode recovers quickly enough, "softness" may not be needed. If, on the other hand, the "soft" diode is too slow, it's still worse than a "hard" fast diode. In most cases I have encountered, the "soft" diodes I tested were, at the end of the day, inferior because they were much slower (think about 60ns instead of 15ns). This means the shoot-through current level (which increases with time, limited by parasitic inductance) was much higher (think about 50A instead of 10A), meaning high losses. And coming out of this huge current peak is not good for EMI, even if it happens somewhat slower - and this softness means you are wasting even more energy during reverse recovery than what the trr value tells you. So, softness is definitely good but only if it doesn't mean compromising on the opposite side of the very same issue.

Sometimes a simple RC snubber is like magic, and not necessarily wasting that much energy. It's best to try out. Unlike the literature and appnotes teach you, I design the RC snubbers by choosing the C value based simply on my power budget available (limited by the resistor case size, or maximum efficiency drop I'm starting to get uncomfortable with), then just test.

Thanks for your wisdom. Your points about "softness" vs. speed is what I'm having hard time digesting in these diode datasheets, particularly with respect to EMI. I might just get a few samples of different diodes and test them out.

I like your approach regarding snubber design. Perhaps we will switch diodes and incorporate a snubber that won't eat up too much power.

Regarding SiC diodes, do they really have zero reverse recovery current/voltage? I'm not familiar with the semiconductor physics of SiC. If so, it seems that SiC would completely eradicate this issue?

Here's one I was looking at today: https://www.digikey.com/product-detail/en/cree-wolfspeed/C3D03060E/C3D03060E-ND/2080273

[TimNJ]

An ferrite bead works well too, is easier to design in and I suspect it has less losses than a snubber. Just make sure the ferrite bead is choosen well when it comes to the frequency response.

The diode is surface mount (TO-252) which makes that tricky. We do have a wire jumper on the top side of the board, in series with the cathode on the bottom side. I have tried a ferrite bead in that location with no change in EMI signature. I will have to look at the frequency response to check if it was suitable.

[TimNJ]

An ferrite bead works well too, is easier to design in and I suspect it has less losses than a snubber. Just make sure the ferrite bead is choosen well when it comes to the frequency response.

+1 this, but with caveat:

You may very well be at high enough power level that you get a Metcal tip rather than an EMI bead.

That is to say, the ferrite bead's losses (to your circuit, it's basically a saturating L || R) are high enough that its temperature rises to Tc (Curie temperature), where L drops off, losses drop off, and it regulates its temperature.  Meanwhile, your EMI spectrum looks good for the first few seconds of operation, then the spikes rise back up to normal levels.  A hot ferrite bead is a useless ferrite bead.

Solution: either reduce loop inductance, increase commutation time (losses go up), or snub the loop inductance.  Which, since reduction is the first priority and you're unable to do that, you now have to increase it, by adding an inductor (not a ferrite bead -- inductors saturate at rated current, ferrite beads saturate at far less!), and damping or clamping that.

Also, have you done the dance with the Y caps?  Sometimes you want one pri-sec or pri-GND, other times you want as little as possible.  Depends on the exact windup of the transformer, and whatever else is farting out noise (heatsink capacitances, proximity, etc.).  I always put in all combinations, in the PCB design, then try permutations until I've found the minimal set and values.

Assuming, of course, a traditional PFC --> converter design.  If N/A, then YMMV.

I suppose if things are very compact, you could also have induction into the supposed-to-be-filtered side, which can really only be solved by reducing loop areas (use 2+ layer construction??) or making it bigger (getting the filtered side away from the noisy side).

Tim

Thank you! I'm a brand spankin' new engineer with a lot to learn, but I am so thankful for people like you who are willing to share their years of experience.

Interesting thought about the ferrite bead situation. I'm not sure if we can really incorporate a ferrite bead due to the diode being a surface mount part.

Where are you suggesting to add an inductor?

The Y-cap dance! No, not really. Right now, it's just two (in series) from pri-sec and two from line-earth and neutral-earth (differential-mode). I suppose investigating the Y-cap situation should be an agenda item.

[TimNJ]

Also, you're right to suspect that a couple more turns on the CMC might not do very much.  That's up in the cutoff range of most CMCs -- that is, when there's a single bulk winding, the turn-turn capacitance dominates.

Try a CMC with fewer turns, and more sections (usually bank-wound like these,
https://www.digikey.com/product-detail/en/epcos-tdk/B82733V2701B001/495-7017-ND/4945303
https://www.digikey.com/product-detail/en/schaffner-emc-inc/EV28-1.5-02-20M/817-1046-ND/1928625
notice each winding is in two sections with a divider between), or higher permeability core (nanocrystalline, i.e.
https://www.mouser.com/ProductDetail/Vacuumschmelze/T60405-R6131-X037?qs=sGAEpiMZZMsVJzu5wKIZCQ7jTXqOmi27XOwRTjSerLI%3d
and such), or managing to squeeze in one more tiny choke that takes over at the high frequency range (unfortunately, these aren't common, and they usually aren't as compact as their values would imply ).

Tim

Thanks for the suggestion. Can you explain the advantage of changing the winding structure of the common-mode choke? We use a separate, small HF common-mode choke on many of our other products but the space constraints dictated here.

[jbb]

A SiC diode would be the modern answer to this problem. Even though more expensive, reducing power losses, getting rid of additional snubber losses, and getting rid of the extra complexity of snubber design and measurement is often worth it.

Regarding SiC diodes, do they really have zero reverse recovery current/voltage? I'm not familiar with the semiconductor physics of SiC. If so, it seems that SiC would completely eradicate this issue?

Here's one I was looking at today: https://www.digikey.com/product-detail/en/cree-wolfspeed/C3D03060E/C3D03060E-ND/2080273

OK, this is definitely a candidate for a SiC diode.  SiC diodes do have zero reverse recovery if:

• They are a Schottky type (which will be the case for basically any 600V or 1200V device).  This means they conduct from N-doped SiC material straight into a metallic terminal.  No holes involved means no need to recombine electrons and holes which causes reverse recovery.
• They are operated within reasonable current rating.  Some SiC Schottkys have a mix of Schottky areas (which do normal duty) and classic PN areas (which only conduct during high current flows).  If the PN areas conduct then they are liable for reverse recovery.  It seems very unlikely for this diode.

Please note that a) the diode still has capacitance and b) higher voltage SiC devices (e.g. 10kV diodes) often (always?) use traditional PN junctions and therefore have reverse recovery.

Comparing your proposed SiC diode to the original part, you may have slightly higher conduction losses (Edit: say 1.2V forward drop vs 0.9V forward drop.  Assuming 0.5A avg forward current, that's 0.6W for SiC P_cond and 0.45W for Si P_cond).  But I would expect your switching losses - in terms of diode reverse recovery and main MOSFET turn-on to drop so you may be better off.  When you consider adding a lossy snubber, I expect you'll be better off overall, and cooler equipment lasts longer :-)

Edit: I'm not sure how fast you're switching: let's take 100 kHz.  For 380V out, the 70nC diode Qrr would be around 26 uCoulomb. At 100 kHz switching, that's 2.6W in the Si diode.  Also, your MOSFET turn on losses will go down because it won't be fighting a reverse recovery current spike.

Depending how far you are into product design, you may find that moving to a SiC diode allows you to push the switching frequency up (without increasing semiconductor losses) and use a smaller, cheaper, lighter inductor.

[T3sl4co1l]

In short, schottky diodes are schottky diodes -- no recovery*, however they do have more capacitance than the other kind, and combined with the sharp grading characteristic of many high voltage diodes and transistors, it can still look a lot like recovery, and cost hard-switching (turn on) losses.

Anecdotally, everywhere I've tried them, they ran cool and worked great.

*Schottky diodes DO still have PN junctions in them, called guard rings -- they are active under reverse breakdown (avalanche) conditions (some diodes are rated for a modest, say ~10mJ, this way), and under heavy forward bias.
 This makes them less desirable for pulse or surge applications, which will still experience reverse recovery.  SiC diodes have high internal resistance, so you may want a larger diode than you expected based on DC losses alone.

Tim

[Circlotron]

2005 was the last time I was involved with the design of continuous mode PFC boost converters (500W to 3kW) and SiC schottkys were only just then becoming available. The difference was like day and night and it was kinda annoying that I had accumulated all these mad skillz with these beautiful lossless passive resonant snubbers and so much of it had become redundant. My vote it use a SiC schottky and see how you go.

[BrianHG]

An ferrite bead works well too, is easier to design in and I suspect it has less losses than a snubber. Just make sure the ferrite bead is choosen well when it comes to the frequency response.

+1 this, but with caveat:

You may very well be at high enough power level that you get a Metcal tip rather than an EMI bead.

That is to say, the ferrite bead's losses (to your circuit, it's basically a saturating L || R) are high enough that its temperature rises to Tc (Curie temperature), where L drops off, losses drop off, and it regulates its temperature.  Meanwhile, your EMI spectrum looks good for the first few seconds of operation, then the spikes rise back up to normal levels.  A hot ferrite bead is a useless ferrite bead.

Solution: either reduce loop inductance, increase commutation time (losses go up), or snub the loop inductance.  Which, since reduction is the first priority and you're unable to do that, you now have to increase it, by adding an inductor (not a ferrite bead -- inductors saturate at rated current, ferrite beads saturate at far less!), and damping or clamping that.

Also, have you done the dance with the Y caps?  Sometimes you want one pri-sec or pri-GND, other times you want as little as possible.  Depends on the exact windup of the transformer, and whatever else is farting out noise (heatsink capacitances, proximity, etc.).  I always put in all combinations, in the PCB design, then try permutations until I've found the minimal set and values.

Assuming, of course, a traditional PFC --> converter design.  If N/A, then YMMV.

I suppose if things are very compact, you could also have induction into the supposed-to-be-filtered side, which can really only be solved by reducing loop areas (use 2+ layer construction??) or making it bigger (getting the filtered side away from the noisy side).

Tim

Thank you! I'm a brand spankin' new engineer with a lot to learn, but I am so thankful for people like you who are willing to share their years of experience.

Interesting thought about the ferrite bead situation. I'm not sure if we can really incorporate a ferrite bead due to the diode being a surface mount part.

Where are you suggesting to add an inductor?

The Y-cap dance! No, not really. Right now, it's just two (in series) from pri-sec and two from line-earth and neutral-earth (differential-mode). I suppose investigating the Y-cap situation should be an agenda item.

SMD ferrite beads exist.  Use them like a series inductor, just mount them adjacent to the diode.
For testing, just lift one side of your diode and attach the smd ferrite bead in series down to the open pad.
You are probably looking for the high current SMD Ferrite chips, which appear as a 0 ohm short at DC, except at their tuned frequency.

To give you an idea, here are the 6amp and above from digikey:
https://www.digikey.com/products/en/filters/ferrite-beads-and-chips/841?FV=114047a%2C114050f%2C1140003%2Cii1%7C2169%2Cffe00349%2C1e0c0014%2C1e0c0019%2C1e0c001a%2C1e0c001c%2C1e0c001d%2C1e0c001f%2C1e0c0023%2C1e0c0027%2C1e0c0028%2C1e0c0043%2C1e0c0060%2C1e0c0061&quantity=&ColumnSort=0&page=1&stock=1&rohs=1&nstock=1&k=ferrite&pageSize=100&pkeyword=ferrite

Remember, you don't need too much impedance up above 10mhz, yet some of these do go into the 100s of ohm range which might be too high for your circuit as it may begin to ring oscillate.  Look as the data sheets where they provide Ohm/Frequency tables.

You wont need the expensive ones, the 2 cent to 10 cent ferrites with something like 25ohm at 100Mhz, 0.005 ohm at DC will fit the bill.

[TimNJ]

2005 was the last time I was involved with the design of continuous mode PFC boost converters (500W to 3kW) and SiC schottkys were only just then becoming available. The difference was like day and night and it was kinda annoying that I had accumulated all these mad skillz with these beautiful lossless passive resonant snubbers and so much of it had become redundant. My vote it use a SiC schottky and see how you go.

Thanks. I ordered a few traditional Si alternatives and an SiC schottky. I'll do some testing today and see how it goes.

[TimNJ]

An ferrite bead works well too, is easier to design in and I suspect it has less losses than a snubber. Just make sure the ferrite bead is choosen well when it comes to the frequency response.

+1 this, but with caveat:

You may very well be at high enough power level that you get a Metcal tip rather than an EMI bead.

That is to say, the ferrite bead's losses (to your circuit, it's basically a saturating L || R) are high enough that its temperature rises to Tc (Curie temperature), where L drops off, losses drop off, and it regulates its temperature.  Meanwhile, your EMI spectrum looks good for the first few seconds of operation, then the spikes rise back up to normal levels.  A hot ferrite bead is a useless ferrite bead.

Solution: either reduce loop inductance, increase commutation time (losses go up), or snub the loop inductance.  Which, since reduction is the first priority and you're unable to do that, you now have to increase it, by adding an inductor (not a ferrite bead -- inductors saturate at rated current, ferrite beads saturate at far less!), and damping or clamping that.

Also, have you done the dance with the Y caps?  Sometimes you want one pri-sec or pri-GND, other times you want as little as possible.  Depends on the exact windup of the transformer, and whatever else is farting out noise (heatsink capacitances, proximity, etc.).  I always put in all combinations, in the PCB design, then try permutations until I've found the minimal set and values.

Assuming, of course, a traditional PFC --> converter design.  If N/A, then YMMV.

I suppose if things are very compact, you could also have induction into the supposed-to-be-filtered side, which can really only be solved by reducing loop areas (use 2+ layer construction??) or making it bigger (getting the filtered side away from the noisy side).

Tim

Thank you! I'm a brand spankin' new engineer with a lot to learn, but I am so thankful for people like you who are willing to share their years of experience.

Interesting thought about the ferrite bead situation. I'm not sure if we can really incorporate a ferrite bead due to the diode being a surface mount part.

Where are you suggesting to add an inductor?

The Y-cap dance! No, not really. Right now, it's just two (in series) from pri-sec and two from line-earth and neutral-earth (differential-mode). I suppose investigating the Y-cap situation should be an agenda item.

SMD ferrite beads exist.  Use them like a series inductor, just mount them adjacent to the diode.
For testing, just lift one side of your diode and attach the smd ferrite bead in series down to the open pad.
You are probably looking for the high current SMD Ferrite chips, which appear as a 0 ohm short at DC, except at their tuned frequency.

To give you an idea, here are the 6amp and above from digikey:
https://www.digikey.com/products/en/filters/ferrite-beads-and-chips/841?FV=114047a%2C114050f%2C1140003%2Cii1%7C2169%2Cffe00349%2C1e0c0014%2C1e0c0019%2C1e0c001a%2C1e0c001c%2C1e0c001d%2C1e0c001f%2C1e0c0023%2C1e0c0027%2C1e0c0028%2C1e0c0043%2C1e0c0060%2C1e0c0061&quantity=&ColumnSort=0&page=1&stock=1&rohs=1&nstock=1&k=ferrite&pageSize=100&pkeyword=ferrite

Remember, you don't need too much impedance up above 10mhz, yet some of these do go into the 100s of ohm range which might be too high for your circuit as it may begin to ring oscillate.  Look as the data sheets where they provide Ohm/Frequency tables.

You wont need the expensive ones, the 2 cent to 10 cent ferrites with something like 25ohm at 100Mhz, 0.005 ohm at DC will fit the bill.

Thanks for the suggestion! Sounds like a good idea. Will try a few out eventually.

25Ohms @ 100MHz is very much in line with what we typically use on TO-220 ferrites. Like this: https://www.mouser.com/ProductDetail/Fair-Rite/2643001501?qs=P8bU7i9nNAXatAT9fBi8Cg%3D%3D